H.264 weighted prediction.
[libav.git] / libavcodec / h264.c
1 /*
2 * H.26L/H.264/AVC/JVT/14496-10/... encoder/decoder
3 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
4 *
5 * This library is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU Lesser General Public
7 * License as published by the Free Software Foundation; either
8 * version 2 of the License, or (at your option) any later version.
9 *
10 * This library is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * Lesser General Public License for more details.
14 *
15 * You should have received a copy of the GNU Lesser General Public
16 * License along with this library; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
18 *
19 */
20
21 /**
22 * @file h264.c
23 * H.264 / AVC / MPEG4 part10 codec.
24 * @author Michael Niedermayer <michaelni@gmx.at>
25 */
26
27 #include "common.h"
28 #include "dsputil.h"
29 #include "avcodec.h"
30 #include "mpegvideo.h"
31 #include "h264data.h"
32 #include "golomb.h"
33
34 #include "cabac.h"
35
36 #undef NDEBUG
37 #include <assert.h>
38
39 #define interlaced_dct interlaced_dct_is_a_bad_name
40 #define mb_intra mb_intra_isnt_initalized_see_mb_type
41
42 #define LUMA_DC_BLOCK_INDEX 25
43 #define CHROMA_DC_BLOCK_INDEX 26
44
45 #define CHROMA_DC_COEFF_TOKEN_VLC_BITS 8
46 #define COEFF_TOKEN_VLC_BITS 8
47 #define TOTAL_ZEROS_VLC_BITS 9
48 #define CHROMA_DC_TOTAL_ZEROS_VLC_BITS 3
49 #define RUN_VLC_BITS 3
50 #define RUN7_VLC_BITS 6
51
52 #define MAX_SPS_COUNT 32
53 #define MAX_PPS_COUNT 256
54
55 #define MAX_MMCO_COUNT 66
56
57 /**
58 * Sequence parameter set
59 */
60 typedef struct SPS{
61
62 int profile_idc;
63 int level_idc;
64 int log2_max_frame_num; ///< log2_max_frame_num_minus4 + 4
65 int poc_type; ///< pic_order_cnt_type
66 int log2_max_poc_lsb; ///< log2_max_pic_order_cnt_lsb_minus4
67 int delta_pic_order_always_zero_flag;
68 int offset_for_non_ref_pic;
69 int offset_for_top_to_bottom_field;
70 int poc_cycle_length; ///< num_ref_frames_in_pic_order_cnt_cycle
71 int ref_frame_count; ///< num_ref_frames
72 int gaps_in_frame_num_allowed_flag;
73 int mb_width; ///< frame_width_in_mbs_minus1 + 1
74 int mb_height; ///< frame_height_in_mbs_minus1 + 1
75 int frame_mbs_only_flag;
76 int mb_aff; ///<mb_adaptive_frame_field_flag
77 int direct_8x8_inference_flag;
78 int crop; ///< frame_cropping_flag
79 int crop_left; ///< frame_cropping_rect_left_offset
80 int crop_right; ///< frame_cropping_rect_right_offset
81 int crop_top; ///< frame_cropping_rect_top_offset
82 int crop_bottom; ///< frame_cropping_rect_bottom_offset
83 int vui_parameters_present_flag;
84 AVRational sar;
85 int timing_info_present_flag;
86 uint32_t num_units_in_tick;
87 uint32_t time_scale;
88 int fixed_frame_rate_flag;
89 short offset_for_ref_frame[256]; //FIXME dyn aloc?
90 }SPS;
91
92 /**
93 * Picture parameter set
94 */
95 typedef struct PPS{
96 int sps_id;
97 int cabac; ///< entropy_coding_mode_flag
98 int pic_order_present; ///< pic_order_present_flag
99 int slice_group_count; ///< num_slice_groups_minus1 + 1
100 int mb_slice_group_map_type;
101 int ref_count[2]; ///< num_ref_idx_l0/1_active_minus1 + 1
102 int weighted_pred; ///< weighted_pred_flag
103 int weighted_bipred_idc;
104 int init_qp; ///< pic_init_qp_minus26 + 26
105 int init_qs; ///< pic_init_qs_minus26 + 26
106 int chroma_qp_index_offset;
107 int deblocking_filter_parameters_present; ///< deblocking_filter_parameters_present_flag
108 int constrained_intra_pred; ///< constrained_intra_pred_flag
109 int redundant_pic_cnt_present; ///< redundant_pic_cnt_present_flag
110 }PPS;
111
112 /**
113 * Memory management control operation opcode.
114 */
115 typedef enum MMCOOpcode{
116 MMCO_END=0,
117 MMCO_SHORT2UNUSED,
118 MMCO_LONG2UNUSED,
119 MMCO_SHORT2LONG,
120 MMCO_SET_MAX_LONG,
121 MMCO_RESET,
122 MMCO_LONG,
123 } MMCOOpcode;
124
125 /**
126 * Memory management control operation.
127 */
128 typedef struct MMCO{
129 MMCOOpcode opcode;
130 int short_frame_num;
131 int long_index;
132 } MMCO;
133
134 /**
135 * H264Context
136 */
137 typedef struct H264Context{
138 MpegEncContext s;
139 int nal_ref_idc;
140 int nal_unit_type;
141 #define NAL_SLICE 1
142 #define NAL_DPA 2
143 #define NAL_DPB 3
144 #define NAL_DPC 4
145 #define NAL_IDR_SLICE 5
146 #define NAL_SEI 6
147 #define NAL_SPS 7
148 #define NAL_PPS 8
149 #define NAL_PICTURE_DELIMITER 9
150 #define NAL_FILTER_DATA 10
151 uint8_t *rbsp_buffer;
152 int rbsp_buffer_size;
153
154 /**
155 * Used to parse AVC variant of h264
156 */
157 int is_avc; ///< this flag is != 0 if codec is avc1
158 int got_avcC; ///< flag used to parse avcC data only once
159 int nal_length_size; ///< Number of bytes used for nal length (1, 2 or 4)
160
161 int chroma_qp; //QPc
162
163 int prev_mb_skiped; //FIXME remove (IMHO not used)
164
165 //prediction stuff
166 int chroma_pred_mode;
167 int intra16x16_pred_mode;
168
169 int8_t intra4x4_pred_mode_cache[5*8];
170 int8_t (*intra4x4_pred_mode)[8];
171 void (*pred4x4 [9+3])(uint8_t *src, uint8_t *topright, int stride);//FIXME move to dsp?
172 void (*pred8x8 [4+3])(uint8_t *src, int stride);
173 void (*pred16x16[4+3])(uint8_t *src, int stride);
174 unsigned int topleft_samples_available;
175 unsigned int top_samples_available;
176 unsigned int topright_samples_available;
177 unsigned int left_samples_available;
178 uint8_t (*top_border)[16+2*8];
179 uint8_t left_border[17+2*9];
180
181 /**
182 * non zero coeff count cache.
183 * is 64 if not available.
184 */
185 uint8_t non_zero_count_cache[6*8];
186 uint8_t (*non_zero_count)[16];
187
188 /**
189 * Motion vector cache.
190 */
191 int16_t mv_cache[2][5*8][2];
192 int8_t ref_cache[2][5*8];
193 #define LIST_NOT_USED -1 //FIXME rename?
194 #define PART_NOT_AVAILABLE -2
195
196 /**
197 * is 1 if the specific list MV&references are set to 0,0,-2.
198 */
199 int mv_cache_clean[2];
200
201 int block_offset[16+8];
202 int chroma_subblock_offset[16]; //FIXME remove
203
204 uint16_t *mb2b_xy; //FIXME are these 4 a good idea?
205 uint16_t *mb2b8_xy;
206 int b_stride; //FIXME use s->b4_stride
207 int b8_stride;
208
209 int halfpel_flag;
210 int thirdpel_flag;
211
212 int unknown_svq3_flag;
213 int next_slice_index;
214
215 SPS sps_buffer[MAX_SPS_COUNT];
216 SPS sps; ///< current sps
217
218 PPS pps_buffer[MAX_PPS_COUNT];
219 /**
220 * current pps
221 */
222 PPS pps; //FIXME move tp Picture perhaps? (->no) do we need that?
223
224 int slice_num;
225 uint8_t *slice_table_base;
226 uint8_t *slice_table; ///< slice_table_base + mb_stride + 1
227 int slice_type;
228 int slice_type_fixed;
229
230 //interlacing specific flags
231 int mb_field_decoding_flag;
232
233 int sub_mb_type[4];
234
235 //POC stuff
236 int poc_lsb;
237 int poc_msb;
238 int delta_poc_bottom;
239 int delta_poc[2];
240 int frame_num;
241 int prev_poc_msb; ///< poc_msb of the last reference pic for POC type 0
242 int prev_poc_lsb; ///< poc_lsb of the last reference pic for POC type 0
243 int frame_num_offset; ///< for POC type 2
244 int prev_frame_num_offset; ///< for POC type 2
245 int prev_frame_num; ///< frame_num of the last pic for POC type 1/2
246
247 /**
248 * frame_num for frames or 2*frame_num for field pics.
249 */
250 int curr_pic_num;
251
252 /**
253 * max_frame_num or 2*max_frame_num for field pics.
254 */
255 int max_pic_num;
256
257 //Weighted pred stuff
258 int use_weight;
259 int use_weight_chroma;
260 int luma_log2_weight_denom;
261 int chroma_log2_weight_denom;
262 int luma_weight[2][16];
263 int luma_offset[2][16];
264 int chroma_weight[2][16][2];
265 int chroma_offset[2][16][2];
266 int implicit_weight[16][16];
267
268 //deblock
269 int deblocking_filter; ///< disable_deblocking_filter_idc with 1<->0
270 int slice_alpha_c0_offset;
271 int slice_beta_offset;
272
273 int redundant_pic_count;
274
275 int direct_spatial_mv_pred;
276 int dist_scale_factor[16];
277
278 /**
279 * num_ref_idx_l0/1_active_minus1 + 1
280 */
281 int ref_count[2];// FIXME split for AFF
282 Picture *short_ref[16];
283 Picture *long_ref[16];
284 Picture default_ref_list[2][32];
285 Picture ref_list[2][32]; //FIXME size?
286 Picture field_ref_list[2][32]; //FIXME size?
287 Picture *delayed_pic[16]; //FIXME size?
288
289 /**
290 * memory management control operations buffer.
291 */
292 MMCO mmco[MAX_MMCO_COUNT];
293 int mmco_index;
294
295 int long_ref_count; ///< number of actual long term references
296 int short_ref_count; ///< number of actual short term references
297
298 //data partitioning
299 GetBitContext intra_gb;
300 GetBitContext inter_gb;
301 GetBitContext *intra_gb_ptr;
302 GetBitContext *inter_gb_ptr;
303
304 DCTELEM mb[16*24] __align8;
305
306 /**
307 * Cabac
308 */
309 CABACContext cabac;
310 uint8_t cabac_state[399];
311 int cabac_init_idc;
312
313 /* 0x100 -> non null luma_dc, 0x80/0x40 -> non null chroma_dc (cb/cr), 0x?0 -> chroma_cbp(0,1,2), 0x0? luma_cbp */
314 uint16_t *cbp_table;
315 int top_cbp;
316 int left_cbp;
317 /* chroma_pred_mode for i4x4 or i16x16, else 0 */
318 uint8_t *chroma_pred_mode_table;
319 int last_qscale_diff;
320 int16_t (*mvd_table[2])[2];
321 int16_t mvd_cache[2][5*8][2];
322 uint8_t *direct_table;
323 uint8_t direct_cache[5*8];
324
325 }H264Context;
326
327 static VLC coeff_token_vlc[4];
328 static VLC chroma_dc_coeff_token_vlc;
329
330 static VLC total_zeros_vlc[15];
331 static VLC chroma_dc_total_zeros_vlc[3];
332
333 static VLC run_vlc[6];
334 static VLC run7_vlc;
335
336 static void svq3_luma_dc_dequant_idct_c(DCTELEM *block, int qp);
337 static void svq3_add_idct_c(uint8_t *dst, DCTELEM *block, int stride, int qp, int dc);
338 static void filter_mb( H264Context *h, int mb_x, int mb_y, uint8_t *img_y, uint8_t *img_cb, uint8_t *img_cr);
339
340 static inline uint32_t pack16to32(int a, int b){
341 #ifdef WORDS_BIGENDIAN
342 return (b&0xFFFF) + (a<<16);
343 #else
344 return (a&0xFFFF) + (b<<16);
345 #endif
346 }
347
348 /**
349 * fill a rectangle.
350 * @param h height of the rectangle, should be a constant
351 * @param w width of the rectangle, should be a constant
352 * @param size the size of val (1 or 4), should be a constant
353 */
354 static inline void fill_rectangle(void *vp, int w, int h, int stride, uint32_t val, int size){ //FIXME ensure this IS inlined
355 uint8_t *p= (uint8_t*)vp;
356 assert(size==1 || size==4);
357
358 w *= size;
359 stride *= size;
360
361 //FIXME check what gcc generates for 64 bit on x86 and possible write a 32 bit ver of it
362 if(w==2 && h==2){
363 *(uint16_t*)(p + 0)=
364 *(uint16_t*)(p + stride)= size==4 ? val : val*0x0101;
365 }else if(w==2 && h==4){
366 *(uint16_t*)(p + 0*stride)=
367 *(uint16_t*)(p + 1*stride)=
368 *(uint16_t*)(p + 2*stride)=
369 *(uint16_t*)(p + 3*stride)= size==4 ? val : val*0x0101;
370 }else if(w==4 && h==1){
371 *(uint32_t*)(p + 0*stride)= size==4 ? val : val*0x01010101;
372 }else if(w==4 && h==2){
373 *(uint32_t*)(p + 0*stride)=
374 *(uint32_t*)(p + 1*stride)= size==4 ? val : val*0x01010101;
375 }else if(w==4 && h==4){
376 *(uint32_t*)(p + 0*stride)=
377 *(uint32_t*)(p + 1*stride)=
378 *(uint32_t*)(p + 2*stride)=
379 *(uint32_t*)(p + 3*stride)= size==4 ? val : val*0x01010101;
380 }else if(w==8 && h==1){
381 *(uint32_t*)(p + 0)=
382 *(uint32_t*)(p + 4)= size==4 ? val : val*0x01010101;
383 }else if(w==8 && h==2){
384 *(uint32_t*)(p + 0 + 0*stride)=
385 *(uint32_t*)(p + 4 + 0*stride)=
386 *(uint32_t*)(p + 0 + 1*stride)=
387 *(uint32_t*)(p + 4 + 1*stride)= size==4 ? val : val*0x01010101;
388 }else if(w==8 && h==4){
389 *(uint64_t*)(p + 0*stride)=
390 *(uint64_t*)(p + 1*stride)=
391 *(uint64_t*)(p + 2*stride)=
392 *(uint64_t*)(p + 3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
393 }else if(w==16 && h==2){
394 *(uint64_t*)(p + 0+0*stride)=
395 *(uint64_t*)(p + 8+0*stride)=
396 *(uint64_t*)(p + 0+1*stride)=
397 *(uint64_t*)(p + 8+1*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
398 }else if(w==16 && h==4){
399 *(uint64_t*)(p + 0+0*stride)=
400 *(uint64_t*)(p + 8+0*stride)=
401 *(uint64_t*)(p + 0+1*stride)=
402 *(uint64_t*)(p + 8+1*stride)=
403 *(uint64_t*)(p + 0+2*stride)=
404 *(uint64_t*)(p + 8+2*stride)=
405 *(uint64_t*)(p + 0+3*stride)=
406 *(uint64_t*)(p + 8+3*stride)= size==4 ? val*0x0100000001ULL : val*0x0101010101010101ULL;
407 }else
408 assert(0);
409 }
410
411 static inline void fill_caches(H264Context *h, int mb_type){
412 MpegEncContext * const s = &h->s;
413 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
414 int topleft_xy, top_xy, topright_xy, left_xy[2];
415 int topleft_type, top_type, topright_type, left_type[2];
416 int left_block[4];
417 int i;
418
419 //wow what a mess, why didnt they simplify the interlacing&intra stuff, i cant imagine that these complex rules are worth it
420
421 if(h->sps.mb_aff){
422 //FIXME
423 topleft_xy = 0; /* avoid warning */
424 top_xy = 0; /* avoid warning */
425 topright_xy = 0; /* avoid warning */
426 }else{
427 topleft_xy = mb_xy-1 - s->mb_stride;
428 top_xy = mb_xy - s->mb_stride;
429 topright_xy= mb_xy+1 - s->mb_stride;
430 left_xy[0] = mb_xy-1;
431 left_xy[1] = mb_xy-1;
432 left_block[0]= 0;
433 left_block[1]= 1;
434 left_block[2]= 2;
435 left_block[3]= 3;
436 }
437
438 topleft_type = h->slice_table[topleft_xy ] == h->slice_num ? s->current_picture.mb_type[topleft_xy] : 0;
439 top_type = h->slice_table[top_xy ] == h->slice_num ? s->current_picture.mb_type[top_xy] : 0;
440 topright_type= h->slice_table[topright_xy] == h->slice_num ? s->current_picture.mb_type[topright_xy]: 0;
441 left_type[0] = h->slice_table[left_xy[0] ] == h->slice_num ? s->current_picture.mb_type[left_xy[0]] : 0;
442 left_type[1] = h->slice_table[left_xy[1] ] == h->slice_num ? s->current_picture.mb_type[left_xy[1]] : 0;
443
444 if(IS_INTRA(mb_type)){
445 h->topleft_samples_available=
446 h->top_samples_available=
447 h->left_samples_available= 0xFFFF;
448 h->topright_samples_available= 0xEEEA;
449
450 if(!IS_INTRA(top_type) && (top_type==0 || h->pps.constrained_intra_pred)){
451 h->topleft_samples_available= 0xB3FF;
452 h->top_samples_available= 0x33FF;
453 h->topright_samples_available= 0x26EA;
454 }
455 for(i=0; i<2; i++){
456 if(!IS_INTRA(left_type[i]) && (left_type[i]==0 || h->pps.constrained_intra_pred)){
457 h->topleft_samples_available&= 0xDF5F;
458 h->left_samples_available&= 0x5F5F;
459 }
460 }
461
462 if(!IS_INTRA(topleft_type) && (topleft_type==0 || h->pps.constrained_intra_pred))
463 h->topleft_samples_available&= 0x7FFF;
464
465 if(!IS_INTRA(topright_type) && (topright_type==0 || h->pps.constrained_intra_pred))
466 h->topright_samples_available&= 0xFBFF;
467
468 if(IS_INTRA4x4(mb_type)){
469 if(IS_INTRA4x4(top_type)){
470 h->intra4x4_pred_mode_cache[4+8*0]= h->intra4x4_pred_mode[top_xy][4];
471 h->intra4x4_pred_mode_cache[5+8*0]= h->intra4x4_pred_mode[top_xy][5];
472 h->intra4x4_pred_mode_cache[6+8*0]= h->intra4x4_pred_mode[top_xy][6];
473 h->intra4x4_pred_mode_cache[7+8*0]= h->intra4x4_pred_mode[top_xy][3];
474 }else{
475 int pred;
476 if(IS_INTRA16x16(top_type) || (IS_INTER(top_type) && !h->pps.constrained_intra_pred))
477 pred= 2;
478 else{
479 pred= -1;
480 }
481 h->intra4x4_pred_mode_cache[4+8*0]=
482 h->intra4x4_pred_mode_cache[5+8*0]=
483 h->intra4x4_pred_mode_cache[6+8*0]=
484 h->intra4x4_pred_mode_cache[7+8*0]= pred;
485 }
486 for(i=0; i<2; i++){
487 if(IS_INTRA4x4(left_type[i])){
488 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[0+2*i]];
489 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= h->intra4x4_pred_mode[left_xy[i]][left_block[1+2*i]];
490 }else{
491 int pred;
492 if(IS_INTRA16x16(left_type[i]) || (IS_INTER(left_type[i]) && !h->pps.constrained_intra_pred))
493 pred= 2;
494 else{
495 pred= -1;
496 }
497 h->intra4x4_pred_mode_cache[3+8*1 + 2*8*i]=
498 h->intra4x4_pred_mode_cache[3+8*2 + 2*8*i]= pred;
499 }
500 }
501 }
502 }
503
504
505 /*
506 0 . T T. T T T T
507 1 L . .L . . . .
508 2 L . .L . . . .
509 3 . T TL . . . .
510 4 L . .L . . . .
511 5 L . .. . . . .
512 */
513 //FIXME constraint_intra_pred & partitioning & nnz (lets hope this is just a typo in the spec)
514 if(top_type){
515 h->non_zero_count_cache[4+8*0]= h->non_zero_count[top_xy][0];
516 h->non_zero_count_cache[5+8*0]= h->non_zero_count[top_xy][1];
517 h->non_zero_count_cache[6+8*0]= h->non_zero_count[top_xy][2];
518 h->non_zero_count_cache[7+8*0]= h->non_zero_count[top_xy][3];
519
520 h->non_zero_count_cache[1+8*0]= h->non_zero_count[top_xy][7];
521 h->non_zero_count_cache[2+8*0]= h->non_zero_count[top_xy][8];
522
523 h->non_zero_count_cache[1+8*3]= h->non_zero_count[top_xy][10];
524 h->non_zero_count_cache[2+8*3]= h->non_zero_count[top_xy][11];
525
526 h->top_cbp= h->cbp_table[top_xy];
527 }else{
528 h->non_zero_count_cache[4+8*0]=
529 h->non_zero_count_cache[5+8*0]=
530 h->non_zero_count_cache[6+8*0]=
531 h->non_zero_count_cache[7+8*0]=
532
533 h->non_zero_count_cache[1+8*0]=
534 h->non_zero_count_cache[2+8*0]=
535
536 h->non_zero_count_cache[1+8*3]=
537 h->non_zero_count_cache[2+8*3]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
538
539 if(IS_INTRA(mb_type)) h->top_cbp= 0x1C0;
540 else h->top_cbp= 0;
541 }
542
543 if(left_type[0]){
544 h->non_zero_count_cache[3+8*1]= h->non_zero_count[left_xy[0]][6];
545 h->non_zero_count_cache[3+8*2]= h->non_zero_count[left_xy[0]][5];
546 h->non_zero_count_cache[0+8*1]= h->non_zero_count[left_xy[0]][9]; //FIXME left_block
547 h->non_zero_count_cache[0+8*4]= h->non_zero_count[left_xy[0]][12];
548 h->left_cbp= h->cbp_table[left_xy[0]]; //FIXME interlacing
549 }else{
550 h->non_zero_count_cache[3+8*1]=
551 h->non_zero_count_cache[3+8*2]=
552 h->non_zero_count_cache[0+8*1]=
553 h->non_zero_count_cache[0+8*4]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
554
555 if(IS_INTRA(mb_type)) h->left_cbp= 0x1C0;//FIXME interlacing
556 else h->left_cbp= 0;
557 }
558
559 if(left_type[1]){
560 h->non_zero_count_cache[3+8*3]= h->non_zero_count[left_xy[1]][4];
561 h->non_zero_count_cache[3+8*4]= h->non_zero_count[left_xy[1]][3];
562 h->non_zero_count_cache[0+8*2]= h->non_zero_count[left_xy[1]][8];
563 h->non_zero_count_cache[0+8*5]= h->non_zero_count[left_xy[1]][11];
564 }else{
565 h->non_zero_count_cache[3+8*3]=
566 h->non_zero_count_cache[3+8*4]=
567 h->non_zero_count_cache[0+8*2]=
568 h->non_zero_count_cache[0+8*5]= h->pps.cabac && !IS_INTRA(mb_type) ? 0 : 64;
569 }
570
571 #if 1
572 //FIXME direct mb can skip much of this
573 if(IS_INTER(mb_type) || (IS_DIRECT(mb_type) && h->direct_spatial_mv_pred)){
574 int list;
575 for(list=0; list<2; list++){
576 if((!IS_8X8(mb_type)) && !USES_LIST(mb_type, list) && !IS_DIRECT(mb_type)){
577 /*if(!h->mv_cache_clean[list]){
578 memset(h->mv_cache [list], 0, 8*5*2*sizeof(int16_t)); //FIXME clean only input? clean at all?
579 memset(h->ref_cache[list], PART_NOT_AVAILABLE, 8*5*sizeof(int8_t));
580 h->mv_cache_clean[list]= 1;
581 }*/
582 continue;
583 }
584 h->mv_cache_clean[list]= 0;
585
586 if(IS_INTER(topleft_type)){
587 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
588 const int b8_xy= h->mb2b8_xy[topleft_xy] + 1 + h->b8_stride;
589 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
590 h->ref_cache[list][scan8[0] - 1 - 1*8]= s->current_picture.ref_index[list][b8_xy];
591 }else{
592 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 - 1*8]= 0;
593 h->ref_cache[list][scan8[0] - 1 - 1*8]= topleft_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
594 }
595
596 if(IS_INTER(top_type)){
597 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
598 const int b8_xy= h->mb2b8_xy[top_xy] + h->b8_stride;
599 *(uint32_t*)h->mv_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 0];
600 *(uint32_t*)h->mv_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 1];
601 *(uint32_t*)h->mv_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 2];
602 *(uint32_t*)h->mv_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + 3];
603 h->ref_cache[list][scan8[0] + 0 - 1*8]=
604 h->ref_cache[list][scan8[0] + 1 - 1*8]= s->current_picture.ref_index[list][b8_xy + 0];
605 h->ref_cache[list][scan8[0] + 2 - 1*8]=
606 h->ref_cache[list][scan8[0] + 3 - 1*8]= s->current_picture.ref_index[list][b8_xy + 1];
607 }else{
608 *(uint32_t*)h->mv_cache [list][scan8[0] + 0 - 1*8]=
609 *(uint32_t*)h->mv_cache [list][scan8[0] + 1 - 1*8]=
610 *(uint32_t*)h->mv_cache [list][scan8[0] + 2 - 1*8]=
611 *(uint32_t*)h->mv_cache [list][scan8[0] + 3 - 1*8]= 0;
612 *(uint32_t*)&h->ref_cache[list][scan8[0] + 0 - 1*8]= ((top_type ? LIST_NOT_USED : PART_NOT_AVAILABLE)&0xFF)*0x01010101;
613 }
614
615 if(IS_INTER(topright_type)){
616 const int b_xy= h->mb2b_xy[topright_xy] + 3*h->b_stride;
617 const int b8_xy= h->mb2b8_xy[topright_xy] + h->b8_stride;
618 *(uint32_t*)h->mv_cache[list][scan8[0] + 4 - 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy];
619 h->ref_cache[list][scan8[0] + 4 - 1*8]= s->current_picture.ref_index[list][b8_xy];
620 }else{
621 *(uint32_t*)h->mv_cache [list][scan8[0] + 4 - 1*8]= 0;
622 h->ref_cache[list][scan8[0] + 4 - 1*8]= topright_type ? LIST_NOT_USED : PART_NOT_AVAILABLE;
623 }
624
625 //FIXME unify cleanup or sth
626 if(IS_INTER(left_type[0])){
627 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
628 const int b8_xy= h->mb2b8_xy[left_xy[0]] + 1;
629 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 + 0*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[0]];
630 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 + 1*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[1]];
631 h->ref_cache[list][scan8[0] - 1 + 0*8]=
632 h->ref_cache[list][scan8[0] - 1 + 1*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[0]>>1)];
633 }else{
634 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 0*8]=
635 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 1*8]= 0;
636 h->ref_cache[list][scan8[0] - 1 + 0*8]=
637 h->ref_cache[list][scan8[0] - 1 + 1*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
638 }
639
640 if(IS_INTER(left_type[1])){
641 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
642 const int b8_xy= h->mb2b8_xy[left_xy[1]] + 1;
643 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 + 2*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[2]];
644 *(uint32_t*)h->mv_cache[list][scan8[0] - 1 + 3*8]= *(uint32_t*)s->current_picture.motion_val[list][b_xy + h->b_stride*left_block[3]];
645 h->ref_cache[list][scan8[0] - 1 + 2*8]=
646 h->ref_cache[list][scan8[0] - 1 + 3*8]= s->current_picture.ref_index[list][b8_xy + h->b8_stride*(left_block[2]>>1)];
647 }else{
648 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 2*8]=
649 *(uint32_t*)h->mv_cache [list][scan8[0] - 1 + 3*8]= 0;
650 h->ref_cache[list][scan8[0] - 1 + 2*8]=
651 h->ref_cache[list][scan8[0] - 1 + 3*8]= left_type[0] ? LIST_NOT_USED : PART_NOT_AVAILABLE;
652 }
653
654 h->ref_cache[list][scan8[5 ]+1] =
655 h->ref_cache[list][scan8[7 ]+1] =
656 h->ref_cache[list][scan8[13]+1] = //FIXME remove past 3 (init somewher else)
657 h->ref_cache[list][scan8[4 ]] =
658 h->ref_cache[list][scan8[12]] = PART_NOT_AVAILABLE;
659 *(uint32_t*)h->mv_cache [list][scan8[5 ]+1]=
660 *(uint32_t*)h->mv_cache [list][scan8[7 ]+1]=
661 *(uint32_t*)h->mv_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewher else)
662 *(uint32_t*)h->mv_cache [list][scan8[4 ]]=
663 *(uint32_t*)h->mv_cache [list][scan8[12]]= 0;
664
665 if( h->pps.cabac ) {
666 /* XXX beurk, Load mvd */
667 if(IS_INTER(topleft_type)){
668 const int b_xy = h->mb2b_xy[topleft_xy] + 3 + 3*h->b_stride;
669 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy];
670 }else{
671 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 - 1*8]= 0;
672 }
673
674 if(IS_INTER(top_type)){
675 const int b_xy= h->mb2b_xy[top_xy] + 3*h->b_stride;
676 *(uint32_t*)h->mvd_cache[list][scan8[0] + 0 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 0];
677 *(uint32_t*)h->mvd_cache[list][scan8[0] + 1 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 1];
678 *(uint32_t*)h->mvd_cache[list][scan8[0] + 2 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 2];
679 *(uint32_t*)h->mvd_cache[list][scan8[0] + 3 - 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + 3];
680 }else{
681 *(uint32_t*)h->mvd_cache [list][scan8[0] + 0 - 1*8]=
682 *(uint32_t*)h->mvd_cache [list][scan8[0] + 1 - 1*8]=
683 *(uint32_t*)h->mvd_cache [list][scan8[0] + 2 - 1*8]=
684 *(uint32_t*)h->mvd_cache [list][scan8[0] + 3 - 1*8]= 0;
685 }
686 if(IS_INTER(left_type[0])){
687 const int b_xy= h->mb2b_xy[left_xy[0]] + 3;
688 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 0*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[0]];
689 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 1*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[1]];
690 }else{
691 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 0*8]=
692 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 1*8]= 0;
693 }
694 if(IS_INTER(left_type[1])){
695 const int b_xy= h->mb2b_xy[left_xy[1]] + 3;
696 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 2*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[2]];
697 *(uint32_t*)h->mvd_cache[list][scan8[0] - 1 + 3*8]= *(uint32_t*)h->mvd_table[list][b_xy + h->b_stride*left_block[3]];
698 }else{
699 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 2*8]=
700 *(uint32_t*)h->mvd_cache [list][scan8[0] - 1 + 3*8]= 0;
701 }
702 *(uint32_t*)h->mvd_cache [list][scan8[5 ]+1]=
703 *(uint32_t*)h->mvd_cache [list][scan8[7 ]+1]=
704 *(uint32_t*)h->mvd_cache [list][scan8[13]+1]= //FIXME remove past 3 (init somewher else)
705 *(uint32_t*)h->mvd_cache [list][scan8[4 ]]=
706 *(uint32_t*)h->mvd_cache [list][scan8[12]]= 0;
707
708 if(h->slice_type == B_TYPE){
709 fill_rectangle(&h->direct_cache[scan8[0]], 4, 4, 8, 0, 1);
710
711 if(IS_DIRECT(top_type)){
712 *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0x01010101;
713 }else if(IS_8X8(top_type)){
714 int b8_xy = h->mb2b8_xy[top_xy] + h->b8_stride;
715 h->direct_cache[scan8[0] + 0 - 1*8]= h->direct_table[b8_xy];
716 h->direct_cache[scan8[0] + 2 - 1*8]= h->direct_table[b8_xy + 1];
717 }else{
718 *(uint32_t*)&h->direct_cache[scan8[0] - 1*8]= 0;
719 }
720
721 //FIXME interlacing
722 if(IS_DIRECT(left_type[0])){
723 h->direct_cache[scan8[0] - 1 + 0*8]=
724 h->direct_cache[scan8[0] - 1 + 2*8]= 1;
725 }else if(IS_8X8(left_type[0])){
726 int b8_xy = h->mb2b8_xy[left_xy[0]] + 1;
727 h->direct_cache[scan8[0] - 1 + 0*8]= h->direct_table[b8_xy];
728 h->direct_cache[scan8[0] - 1 + 2*8]= h->direct_table[b8_xy + h->b8_stride];
729 }else{
730 h->direct_cache[scan8[0] - 1 + 0*8]=
731 h->direct_cache[scan8[0] - 1 + 2*8]= 0;
732 }
733 }
734 }
735 }
736 }
737 #endif
738 }
739
740 static inline void write_back_intra_pred_mode(H264Context *h){
741 MpegEncContext * const s = &h->s;
742 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
743
744 h->intra4x4_pred_mode[mb_xy][0]= h->intra4x4_pred_mode_cache[7+8*1];
745 h->intra4x4_pred_mode[mb_xy][1]= h->intra4x4_pred_mode_cache[7+8*2];
746 h->intra4x4_pred_mode[mb_xy][2]= h->intra4x4_pred_mode_cache[7+8*3];
747 h->intra4x4_pred_mode[mb_xy][3]= h->intra4x4_pred_mode_cache[7+8*4];
748 h->intra4x4_pred_mode[mb_xy][4]= h->intra4x4_pred_mode_cache[4+8*4];
749 h->intra4x4_pred_mode[mb_xy][5]= h->intra4x4_pred_mode_cache[5+8*4];
750 h->intra4x4_pred_mode[mb_xy][6]= h->intra4x4_pred_mode_cache[6+8*4];
751 }
752
753 /**
754 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
755 */
756 static inline int check_intra4x4_pred_mode(H264Context *h){
757 MpegEncContext * const s = &h->s;
758 static const int8_t top [12]= {-1, 0,LEFT_DC_PRED,-1,-1,-1,-1,-1, 0};
759 static const int8_t left[12]= { 0,-1, TOP_DC_PRED, 0,-1,-1,-1, 0,-1,DC_128_PRED};
760 int i;
761
762 if(!(h->top_samples_available&0x8000)){
763 for(i=0; i<4; i++){
764 int status= top[ h->intra4x4_pred_mode_cache[scan8[0] + i] ];
765 if(status<0){
766 av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
767 return -1;
768 } else if(status){
769 h->intra4x4_pred_mode_cache[scan8[0] + i]= status;
770 }
771 }
772 }
773
774 if(!(h->left_samples_available&0x8000)){
775 for(i=0; i<4; i++){
776 int status= left[ h->intra4x4_pred_mode_cache[scan8[0] + 8*i] ];
777 if(status<0){
778 av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra4x4 mode %d at %d %d\n", status, s->mb_x, s->mb_y);
779 return -1;
780 } else if(status){
781 h->intra4x4_pred_mode_cache[scan8[0] + 8*i]= status;
782 }
783 }
784 }
785
786 return 0;
787 } //FIXME cleanup like next
788
789 /**
790 * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
791 */
792 static inline int check_intra_pred_mode(H264Context *h, int mode){
793 MpegEncContext * const s = &h->s;
794 static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
795 static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
796
797 if(mode < 0 || mode > 6) {
798 av_log(h->s.avctx, AV_LOG_ERROR, "out of range intra chroma pred mode at %d %d\n", s->mb_x, s->mb_y);
799 return -1;
800 }
801
802 if(!(h->top_samples_available&0x8000)){
803 mode= top[ mode ];
804 if(mode<0){
805 av_log(h->s.avctx, AV_LOG_ERROR, "top block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
806 return -1;
807 }
808 }
809
810 if(!(h->left_samples_available&0x8000)){
811 mode= left[ mode ];
812 if(mode<0){
813 av_log(h->s.avctx, AV_LOG_ERROR, "left block unavailable for requested intra mode at %d %d\n", s->mb_x, s->mb_y);
814 return -1;
815 }
816 }
817
818 return mode;
819 }
820
821 /**
822 * gets the predicted intra4x4 prediction mode.
823 */
824 static inline int pred_intra_mode(H264Context *h, int n){
825 const int index8= scan8[n];
826 const int left= h->intra4x4_pred_mode_cache[index8 - 1];
827 const int top = h->intra4x4_pred_mode_cache[index8 - 8];
828 const int min= FFMIN(left, top);
829
830 tprintf("mode:%d %d min:%d\n", left ,top, min);
831
832 if(min<0) return DC_PRED;
833 else return min;
834 }
835
836 static inline void write_back_non_zero_count(H264Context *h){
837 MpegEncContext * const s = &h->s;
838 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
839
840 h->non_zero_count[mb_xy][0]= h->non_zero_count_cache[4+8*4];
841 h->non_zero_count[mb_xy][1]= h->non_zero_count_cache[5+8*4];
842 h->non_zero_count[mb_xy][2]= h->non_zero_count_cache[6+8*4];
843 h->non_zero_count[mb_xy][3]= h->non_zero_count_cache[7+8*4];
844 h->non_zero_count[mb_xy][4]= h->non_zero_count_cache[7+8*3];
845 h->non_zero_count[mb_xy][5]= h->non_zero_count_cache[7+8*2];
846 h->non_zero_count[mb_xy][6]= h->non_zero_count_cache[7+8*1];
847
848 h->non_zero_count[mb_xy][7]= h->non_zero_count_cache[1+8*2];
849 h->non_zero_count[mb_xy][8]= h->non_zero_count_cache[2+8*2];
850 h->non_zero_count[mb_xy][9]= h->non_zero_count_cache[2+8*1];
851
852 h->non_zero_count[mb_xy][10]=h->non_zero_count_cache[1+8*5];
853 h->non_zero_count[mb_xy][11]=h->non_zero_count_cache[2+8*5];
854 h->non_zero_count[mb_xy][12]=h->non_zero_count_cache[2+8*4];
855 }
856
857 /**
858 * gets the predicted number of non zero coefficients.
859 * @param n block index
860 */
861 static inline int pred_non_zero_count(H264Context *h, int n){
862 const int index8= scan8[n];
863 const int left= h->non_zero_count_cache[index8 - 1];
864 const int top = h->non_zero_count_cache[index8 - 8];
865 int i= left + top;
866
867 if(i<64) i= (i+1)>>1;
868
869 tprintf("pred_nnz L%X T%X n%d s%d P%X\n", left, top, n, scan8[n], i&31);
870
871 return i&31;
872 }
873
874 static inline int fetch_diagonal_mv(H264Context *h, const int16_t **C, int i, int list, int part_width){
875 const int topright_ref= h->ref_cache[list][ i - 8 + part_width ];
876
877 if(topright_ref != PART_NOT_AVAILABLE){
878 *C= h->mv_cache[list][ i - 8 + part_width ];
879 return topright_ref;
880 }else{
881 tprintf("topright MV not available\n");
882
883 *C= h->mv_cache[list][ i - 8 - 1 ];
884 return h->ref_cache[list][ i - 8 - 1 ];
885 }
886 }
887
888 /**
889 * gets the predicted MV.
890 * @param n the block index
891 * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)
892 * @param mx the x component of the predicted motion vector
893 * @param my the y component of the predicted motion vector
894 */
895 static inline void pred_motion(H264Context * const h, int n, int part_width, int list, int ref, int * const mx, int * const my){
896 const int index8= scan8[n];
897 const int top_ref= h->ref_cache[list][ index8 - 8 ];
898 const int left_ref= h->ref_cache[list][ index8 - 1 ];
899 const int16_t * const A= h->mv_cache[list][ index8 - 1 ];
900 const int16_t * const B= h->mv_cache[list][ index8 - 8 ];
901 const int16_t * C;
902 int diagonal_ref, match_count;
903
904 assert(part_width==1 || part_width==2 || part_width==4);
905
906 /* mv_cache
907 B . . A T T T T
908 U . . L . . , .
909 U . . L . . . .
910 U . . L . . , .
911 . . . L . . . .
912 */
913
914 diagonal_ref= fetch_diagonal_mv(h, &C, index8, list, part_width);
915 match_count= (diagonal_ref==ref) + (top_ref==ref) + (left_ref==ref);
916 if(match_count > 1){ //most common
917 *mx= mid_pred(A[0], B[0], C[0]);
918 *my= mid_pred(A[1], B[1], C[1]);
919 }else if(match_count==1){
920 if(left_ref==ref){
921 *mx= A[0];
922 *my= A[1];
923 }else if(top_ref==ref){
924 *mx= B[0];
925 *my= B[1];
926 }else{
927 *mx= C[0];
928 *my= C[1];
929 }
930 }else{
931 if(top_ref == PART_NOT_AVAILABLE && diagonal_ref == PART_NOT_AVAILABLE && left_ref != PART_NOT_AVAILABLE){
932 *mx= A[0];
933 *my= A[1];
934 }else{
935 *mx= mid_pred(A[0], B[0], C[0]);
936 *my= mid_pred(A[1], B[1], C[1]);
937 }
938 }
939
940 tprintf("pred_motion (%2d %2d %2d) (%2d %2d %2d) (%2d %2d %2d) -> (%2d %2d %2d) at %2d %2d %d list %d\n", top_ref, B[0], B[1], diagonal_ref, C[0], C[1], left_ref, A[0], A[1], ref, *mx, *my, h->s.mb_x, h->s.mb_y, n, list);
941 }
942
943 /**
944 * gets the directionally predicted 16x8 MV.
945 * @param n the block index
946 * @param mx the x component of the predicted motion vector
947 * @param my the y component of the predicted motion vector
948 */
949 static inline void pred_16x8_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
950 if(n==0){
951 const int top_ref= h->ref_cache[list][ scan8[0] - 8 ];
952 const int16_t * const B= h->mv_cache[list][ scan8[0] - 8 ];
953
954 tprintf("pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d\n", top_ref, B[0], B[1], h->s.mb_x, h->s.mb_y, n, list);
955
956 if(top_ref == ref){
957 *mx= B[0];
958 *my= B[1];
959 return;
960 }
961 }else{
962 const int left_ref= h->ref_cache[list][ scan8[8] - 1 ];
963 const int16_t * const A= h->mv_cache[list][ scan8[8] - 1 ];
964
965 tprintf("pred_16x8: (%2d %2d %2d) at %2d %2d %d list %d\n", left_ref, A[0], A[1], h->s.mb_x, h->s.mb_y, n, list);
966
967 if(left_ref == ref){
968 *mx= A[0];
969 *my= A[1];
970 return;
971 }
972 }
973
974 //RARE
975 pred_motion(h, n, 4, list, ref, mx, my);
976 }
977
978 /**
979 * gets the directionally predicted 8x16 MV.
980 * @param n the block index
981 * @param mx the x component of the predicted motion vector
982 * @param my the y component of the predicted motion vector
983 */
984 static inline void pred_8x16_motion(H264Context * const h, int n, int list, int ref, int * const mx, int * const my){
985 if(n==0){
986 const int left_ref= h->ref_cache[list][ scan8[0] - 1 ];
987 const int16_t * const A= h->mv_cache[list][ scan8[0] - 1 ];
988
989 tprintf("pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d\n", left_ref, A[0], A[1], h->s.mb_x, h->s.mb_y, n, list);
990
991 if(left_ref == ref){
992 *mx= A[0];
993 *my= A[1];
994 return;
995 }
996 }else{
997 const int16_t * C;
998 int diagonal_ref;
999
1000 diagonal_ref= fetch_diagonal_mv(h, &C, scan8[4], list, 2);
1001
1002 tprintf("pred_8x16: (%2d %2d %2d) at %2d %2d %d list %d\n", diagonal_ref, C[0], C[1], h->s.mb_x, h->s.mb_y, n, list);
1003
1004 if(diagonal_ref == ref){
1005 *mx= C[0];
1006 *my= C[1];
1007 return;
1008 }
1009 }
1010
1011 //RARE
1012 pred_motion(h, n, 2, list, ref, mx, my);
1013 }
1014
1015 static inline void pred_pskip_motion(H264Context * const h, int * const mx, int * const my){
1016 const int top_ref = h->ref_cache[0][ scan8[0] - 8 ];
1017 const int left_ref= h->ref_cache[0][ scan8[0] - 1 ];
1018
1019 tprintf("pred_pskip: (%d) (%d) at %2d %2d\n", top_ref, left_ref, h->s.mb_x, h->s.mb_y);
1020
1021 if(top_ref == PART_NOT_AVAILABLE || left_ref == PART_NOT_AVAILABLE
1022 || (top_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 8 ] == 0)
1023 || (left_ref == 0 && *(uint32_t*)h->mv_cache[0][ scan8[0] - 1 ] == 0)){
1024
1025 *mx = *my = 0;
1026 return;
1027 }
1028
1029 pred_motion(h, 0, 4, 0, 0, mx, my);
1030
1031 return;
1032 }
1033
1034 static inline void direct_dist_scale_factor(H264Context * const h){
1035 const int poc = h->s.current_picture_ptr->poc;
1036 const int poc1 = h->ref_list[1][0].poc;
1037 int i;
1038 for(i=0; i<h->ref_count[0]; i++){
1039 int poc0 = h->ref_list[0][i].poc;
1040 int td = clip(poc1 - poc0, -128, 127);
1041 if(td == 0 /* FIXME || pic0 is a long-term ref */){
1042 h->dist_scale_factor[i] = 256;
1043 }else{
1044 int tb = clip(poc - poc0, -128, 127);
1045 int tx = (16384 + (ABS(td) >> 1)) / td;
1046 h->dist_scale_factor[i] = clip((tb*tx + 32) >> 6, -1024, 1023);
1047 }
1048 }
1049 }
1050
1051 static inline void pred_direct_motion(H264Context * const h, int *mb_type){
1052 MpegEncContext * const s = &h->s;
1053 const int mb_xy = s->mb_x + s->mb_y*s->mb_stride;
1054 const int b8_xy = 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1055 const int b4_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1056 const int mb_type_col = h->ref_list[1][0].mb_type[mb_xy];
1057 const int16_t (*l1mv0)[2] = (const int16_t (*)[2]) &h->ref_list[1][0].motion_val[0][b4_xy];
1058 const int8_t *l1ref0 = &h->ref_list[1][0].ref_index[0][b8_xy];
1059 const int is_b8x8 = IS_8X8(*mb_type);
1060 int sub_mb_type;
1061 int i8, i4;
1062
1063 if(IS_8X8(mb_type_col) && !h->sps.direct_8x8_inference_flag){
1064 /* FIXME save sub mb types from previous frames (or derive from MVs)
1065 * so we know exactly what block size to use */
1066 sub_mb_type = MB_TYPE_8x8|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_4x4 */
1067 *mb_type = MB_TYPE_8x8;
1068 }else if(!is_b8x8 && (IS_16X16(mb_type_col) || IS_INTRA(mb_type_col))){
1069 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1070 *mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_16x16 */
1071 }else{
1072 sub_mb_type = MB_TYPE_16x16|MB_TYPE_P0L0|MB_TYPE_P0L1|MB_TYPE_DIRECT2; /* B_SUB_8x8 */
1073 *mb_type = MB_TYPE_8x8;
1074 }
1075 if(!is_b8x8)
1076 *mb_type |= MB_TYPE_DIRECT2;
1077
1078 if(h->direct_spatial_mv_pred){
1079 int ref[2];
1080 int mv[2][2];
1081 int list;
1082
1083 /* ref = min(neighbors) */
1084 for(list=0; list<2; list++){
1085 int refa = h->ref_cache[list][scan8[0] - 1];
1086 int refb = h->ref_cache[list][scan8[0] - 8];
1087 int refc = h->ref_cache[list][scan8[0] - 8 + 4];
1088 if(refc == -2)
1089 refc = h->ref_cache[list][scan8[0] - 8 - 1];
1090 ref[list] = refa;
1091 if(ref[list] < 0 || (refb < ref[list] && refb >= 0))
1092 ref[list] = refb;
1093 if(ref[list] < 0 || (refc < ref[list] && refc >= 0))
1094 ref[list] = refc;
1095 if(ref[list] < 0)
1096 ref[list] = -1;
1097 }
1098
1099 if(ref[0] < 0 && ref[1] < 0){
1100 ref[0] = ref[1] = 0;
1101 mv[0][0] = mv[0][1] =
1102 mv[1][0] = mv[1][1] = 0;
1103 }else{
1104 for(list=0; list<2; list++){
1105 if(ref[list] >= 0)
1106 pred_motion(h, 0, 4, list, ref[list], &mv[list][0], &mv[list][1]);
1107 else
1108 mv[list][0] = mv[list][1] = 0;
1109 }
1110 }
1111
1112 if(ref[1] < 0){
1113 *mb_type &= ~MB_TYPE_P0L1;
1114 sub_mb_type &= ~MB_TYPE_P0L1;
1115 }else if(ref[0] < 0){
1116 *mb_type &= ~MB_TYPE_P0L0;
1117 sub_mb_type &= ~MB_TYPE_P0L0;
1118 }
1119
1120 if(IS_16X16(*mb_type)){
1121 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref[0], 1);
1122 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, ref[1], 1);
1123 if(!IS_INTRA(mb_type_col) && l1ref0[0] == 0 &&
1124 ABS(l1mv0[0][0]) <= 1 && ABS(l1mv0[0][1]) <= 1){
1125 if(ref[0] > 0)
1126 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1127 else
1128 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1129 if(ref[1] > 0)
1130 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1131 else
1132 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1133 }else{
1134 fill_rectangle(&h->mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1135 fill_rectangle(&h->mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1136 }
1137 }else{
1138 for(i8=0; i8<4; i8++){
1139 const int x8 = i8&1;
1140 const int y8 = i8>>1;
1141
1142 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1143 continue;
1144 h->sub_mb_type[i8] = sub_mb_type;
1145
1146 fill_rectangle(&h->mv_cache[0][scan8[i8*4]], 2, 2, 8, pack16to32(mv[0][0],mv[0][1]), 4);
1147 fill_rectangle(&h->mv_cache[1][scan8[i8*4]], 2, 2, 8, pack16to32(mv[1][0],mv[1][1]), 4);
1148 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref[0], 1);
1149 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, ref[1], 1);
1150
1151 /* col_zero_flag */
1152 if(!IS_INTRA(mb_type_col) && l1ref0[x8 + y8*h->b8_stride] == 0){
1153 for(i4=0; i4<4; i4++){
1154 const int16_t *mv_col = l1mv0[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1155 if(ABS(mv_col[0]) <= 1 && ABS(mv_col[1]) <= 1){
1156 if(ref[0] == 0)
1157 *(uint32_t*)h->mv_cache[0][scan8[i8*4+i4]] = 0;
1158 if(ref[1] == 0)
1159 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] = 0;
1160 }
1161 }
1162 }
1163 }
1164 }
1165 }else{ /* direct temporal mv pred */
1166 /* FIXME assumes that L1ref0 used the same ref lists as current frame */
1167 if(IS_16X16(*mb_type)){
1168 fill_rectangle(&h->ref_cache[1][scan8[0]], 4, 4, 8, 0, 1);
1169 if(IS_INTRA(mb_type_col)){
1170 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, 0, 1);
1171 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, 0, 4);
1172 fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, 0, 4);
1173 }else{
1174 const int ref0 = l1ref0[0];
1175 const int dist_scale_factor = h->dist_scale_factor[ref0];
1176 const int16_t *mv_col = l1mv0[0];
1177 int mv_l0[2];
1178 mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
1179 mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
1180 fill_rectangle(&h->ref_cache[0][scan8[0]], 4, 4, 8, ref0, 1);
1181 fill_rectangle(&h-> mv_cache[0][scan8[0]], 4, 4, 8, pack16to32(mv_l0[0],mv_l0[1]), 4);
1182 fill_rectangle(&h-> mv_cache[1][scan8[0]], 4, 4, 8, pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]), 4);
1183 }
1184 }else{
1185 for(i8=0; i8<4; i8++){
1186 const int x8 = i8&1;
1187 const int y8 = i8>>1;
1188 int ref0, dist_scale_factor;
1189
1190 if(is_b8x8 && !IS_DIRECT(h->sub_mb_type[i8]))
1191 continue;
1192 h->sub_mb_type[i8] = sub_mb_type;
1193 if(IS_INTRA(mb_type_col)){
1194 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, 0, 1);
1195 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1196 fill_rectangle(&h-> mv_cache[0][scan8[i8*4]], 2, 2, 8, 0, 4);
1197 fill_rectangle(&h-> mv_cache[1][scan8[i8*4]], 2, 2, 8, 0, 4);
1198 continue;
1199 }
1200
1201 ref0 = l1ref0[x8 + y8*h->b8_stride];
1202 dist_scale_factor = h->dist_scale_factor[ref0];
1203
1204 fill_rectangle(&h->ref_cache[0][scan8[i8*4]], 2, 2, 8, ref0, 1);
1205 fill_rectangle(&h->ref_cache[1][scan8[i8*4]], 2, 2, 8, 0, 1);
1206 for(i4=0; i4<4; i4++){
1207 const int16_t *mv_col = l1mv0[x8*2 + (i4&1) + (y8*2 + (i4>>1))*h->b_stride];
1208 int16_t *mv_l0 = h->mv_cache[0][scan8[i8*4+i4]];
1209 mv_l0[0] = (dist_scale_factor * mv_col[0] + 128) >> 8;
1210 mv_l0[1] = (dist_scale_factor * mv_col[1] + 128) >> 8;
1211 *(uint32_t*)h->mv_cache[1][scan8[i8*4+i4]] =
1212 pack16to32(mv_l0[0]-mv_col[0],mv_l0[1]-mv_col[1]);
1213 }
1214 }
1215 }
1216 }
1217 }
1218
1219 static inline void write_back_motion(H264Context *h, int mb_type){
1220 MpegEncContext * const s = &h->s;
1221 const int b_xy = 4*s->mb_x + 4*s->mb_y*h->b_stride;
1222 const int b8_xy= 2*s->mb_x + 2*s->mb_y*h->b8_stride;
1223 int list;
1224
1225 for(list=0; list<2; list++){
1226 int y;
1227 if((!IS_8X8(mb_type)) && !USES_LIST(mb_type, list)){
1228 if(1){ //FIXME skip or never read if mb_type doesnt use it
1229 for(y=0; y<4; y++){
1230 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]=
1231 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= 0;
1232 }
1233 if( h->pps.cabac ) {
1234 /* FIXME needed ? */
1235 for(y=0; y<4; y++){
1236 *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]=
1237 *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= 0;
1238 }
1239 }
1240 for(y=0; y<2; y++){
1241 *(uint16_t*)&s->current_picture.ref_index[list][b8_xy + y*h->b8_stride]= (LIST_NOT_USED&0xFF)*0x0101;
1242 }
1243 }
1244 continue;
1245 }
1246
1247 for(y=0; y<4; y++){
1248 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mv_cache[list][scan8[0]+0 + 8*y];
1249 *(uint64_t*)s->current_picture.motion_val[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mv_cache[list][scan8[0]+2 + 8*y];
1250 }
1251 if( h->pps.cabac ) {
1252 for(y=0; y<4; y++){
1253 *(uint64_t*)h->mvd_table[list][b_xy + 0 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+0 + 8*y];
1254 *(uint64_t*)h->mvd_table[list][b_xy + 2 + y*h->b_stride]= *(uint64_t*)h->mvd_cache[list][scan8[0]+2 + 8*y];
1255 }
1256 }
1257 for(y=0; y<2; y++){
1258 s->current_picture.ref_index[list][b8_xy + 0 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+0 + 16*y];
1259 s->current_picture.ref_index[list][b8_xy + 1 + y*h->b8_stride]= h->ref_cache[list][scan8[0]+2 + 16*y];
1260 }
1261 }
1262
1263 if(h->slice_type == B_TYPE && h->pps.cabac){
1264 if(IS_8X8(mb_type)){
1265 h->direct_table[b8_xy+1+0*h->b8_stride] = IS_DIRECT(h->sub_mb_type[1]) ? 1 : 0;
1266 h->direct_table[b8_xy+0+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[2]) ? 1 : 0;
1267 h->direct_table[b8_xy+1+1*h->b8_stride] = IS_DIRECT(h->sub_mb_type[3]) ? 1 : 0;
1268 }
1269 }
1270 }
1271
1272 /**
1273 * Decodes a network abstraction layer unit.
1274 * @param consumed is the number of bytes used as input
1275 * @param length is the length of the array
1276 * @param dst_length is the number of decoded bytes FIXME here or a decode rbsp ttailing?
1277 * @returns decoded bytes, might be src+1 if no escapes
1278 */
1279 static uint8_t *decode_nal(H264Context *h, uint8_t *src, int *dst_length, int *consumed, int length){
1280 int i, si, di;
1281 uint8_t *dst;
1282
1283 // src[0]&0x80; //forbidden bit
1284 h->nal_ref_idc= src[0]>>5;
1285 h->nal_unit_type= src[0]&0x1F;
1286
1287 src++; length--;
1288 #if 0
1289 for(i=0; i<length; i++)
1290 printf("%2X ", src[i]);
1291 #endif
1292 for(i=0; i+1<length; i+=2){
1293 if(src[i]) continue;
1294 if(i>0 && src[i-1]==0) i--;
1295 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1296 if(src[i+2]!=3){
1297 /* startcode, so we must be past the end */
1298 length=i;
1299 }
1300 break;
1301 }
1302 }
1303
1304 if(i>=length-1){ //no escaped 0
1305 *dst_length= length;
1306 *consumed= length+1; //+1 for the header
1307 return src;
1308 }
1309
1310 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length);
1311 dst= h->rbsp_buffer;
1312
1313 //printf("deoding esc\n");
1314 si=di=0;
1315 while(si<length){
1316 //remove escapes (very rare 1:2^22)
1317 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1318 if(src[si+2]==3){ //escape
1319 dst[di++]= 0;
1320 dst[di++]= 0;
1321 si+=3;
1322 continue;
1323 }else //next start code
1324 break;
1325 }
1326
1327 dst[di++]= src[si++];
1328 }
1329
1330 *dst_length= di;
1331 *consumed= si + 1;//+1 for the header
1332 //FIXME store exact number of bits in the getbitcontext (its needed for decoding)
1333 return dst;
1334 }
1335
1336 #if 0
1337 /**
1338 * @param src the data which should be escaped
1339 * @param dst the target buffer, dst+1 == src is allowed as a special case
1340 * @param length the length of the src data
1341 * @param dst_length the length of the dst array
1342 * @returns length of escaped data in bytes or -1 if an error occured
1343 */
1344 static int encode_nal(H264Context *h, uint8_t *dst, uint8_t *src, int length, int dst_length){
1345 int i, escape_count, si, di;
1346 uint8_t *temp;
1347
1348 assert(length>=0);
1349 assert(dst_length>0);
1350
1351 dst[0]= (h->nal_ref_idc<<5) + h->nal_unit_type;
1352
1353 if(length==0) return 1;
1354
1355 escape_count= 0;
1356 for(i=0; i<length; i+=2){
1357 if(src[i]) continue;
1358 if(i>0 && src[i-1]==0)
1359 i--;
1360 if(i+2<length && src[i+1]==0 && src[i+2]<=3){
1361 escape_count++;
1362 i+=2;
1363 }
1364 }
1365
1366 if(escape_count==0){
1367 if(dst+1 != src)
1368 memcpy(dst+1, src, length);
1369 return length + 1;
1370 }
1371
1372 if(length + escape_count + 1> dst_length)
1373 return -1;
1374
1375 //this should be damn rare (hopefully)
1376
1377 h->rbsp_buffer= av_fast_realloc(h->rbsp_buffer, &h->rbsp_buffer_size, length + escape_count);
1378 temp= h->rbsp_buffer;
1379 //printf("encoding esc\n");
1380
1381 si= 0;
1382 di= 0;
1383 while(si < length){
1384 if(si+2<length && src[si]==0 && src[si+1]==0 && src[si+2]<=3){
1385 temp[di++]= 0; si++;
1386 temp[di++]= 0; si++;
1387 temp[di++]= 3;
1388 temp[di++]= src[si++];
1389 }
1390 else
1391 temp[di++]= src[si++];
1392 }
1393 memcpy(dst+1, temp, length+escape_count);
1394
1395 assert(di == length+escape_count);
1396
1397 return di + 1;
1398 }
1399
1400 /**
1401 * write 1,10,100,1000,... for alignment, yes its exactly inverse to mpeg4
1402 */
1403 static void encode_rbsp_trailing(PutBitContext *pb){
1404 int length;
1405 put_bits(pb, 1, 1);
1406 length= (-put_bits_count(pb))&7;
1407 if(length) put_bits(pb, length, 0);
1408 }
1409 #endif
1410
1411 /**
1412 * identifies the exact end of the bitstream
1413 * @return the length of the trailing, or 0 if damaged
1414 */
1415 static int decode_rbsp_trailing(uint8_t *src){
1416 int v= *src;
1417 int r;
1418
1419 tprintf("rbsp trailing %X\n", v);
1420
1421 for(r=1; r<9; r++){
1422 if(v&1) return r;
1423 v>>=1;
1424 }
1425 return 0;
1426 }
1427
1428 /**
1429 * idct tranforms the 16 dc values and dequantize them.
1430 * @param qp quantization parameter
1431 */
1432 static void h264_luma_dc_dequant_idct_c(DCTELEM *block, int qp){
1433 const int qmul= dequant_coeff[qp][0];
1434 #define stride 16
1435 int i;
1436 int temp[16]; //FIXME check if this is a good idea
1437 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1438 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1439
1440 //memset(block, 64, 2*256);
1441 //return;
1442 for(i=0; i<4; i++){
1443 const int offset= y_offset[i];
1444 const int z0= block[offset+stride*0] + block[offset+stride*4];
1445 const int z1= block[offset+stride*0] - block[offset+stride*4];
1446 const int z2= block[offset+stride*1] - block[offset+stride*5];
1447 const int z3= block[offset+stride*1] + block[offset+stride*5];
1448
1449 temp[4*i+0]= z0+z3;
1450 temp[4*i+1]= z1+z2;
1451 temp[4*i+2]= z1-z2;
1452 temp[4*i+3]= z0-z3;
1453 }
1454
1455 for(i=0; i<4; i++){
1456 const int offset= x_offset[i];
1457 const int z0= temp[4*0+i] + temp[4*2+i];
1458 const int z1= temp[4*0+i] - temp[4*2+i];
1459 const int z2= temp[4*1+i] - temp[4*3+i];
1460 const int z3= temp[4*1+i] + temp[4*3+i];
1461
1462 block[stride*0 +offset]= ((z0 + z3)*qmul + 2)>>2; //FIXME think about merging this into decode_resdual
1463 block[stride*2 +offset]= ((z1 + z2)*qmul + 2)>>2;
1464 block[stride*8 +offset]= ((z1 - z2)*qmul + 2)>>2;
1465 block[stride*10+offset]= ((z0 - z3)*qmul + 2)>>2;
1466 }
1467 }
1468
1469 #if 0
1470 /**
1471 * dct tranforms the 16 dc values.
1472 * @param qp quantization parameter ??? FIXME
1473 */
1474 static void h264_luma_dc_dct_c(DCTELEM *block/*, int qp*/){
1475 // const int qmul= dequant_coeff[qp][0];
1476 int i;
1477 int temp[16]; //FIXME check if this is a good idea
1478 static const int x_offset[4]={0, 1*stride, 4* stride, 5*stride};
1479 static const int y_offset[4]={0, 2*stride, 8* stride, 10*stride};
1480
1481 for(i=0; i<4; i++){
1482 const int offset= y_offset[i];
1483 const int z0= block[offset+stride*0] + block[offset+stride*4];
1484 const int z1= block[offset+stride*0] - block[offset+stride*4];
1485 const int z2= block[offset+stride*1] - block[offset+stride*5];
1486 const int z3= block[offset+stride*1] + block[offset+stride*5];
1487
1488 temp[4*i+0]= z0+z3;
1489 temp[4*i+1]= z1+z2;
1490 temp[4*i+2]= z1-z2;
1491 temp[4*i+3]= z0-z3;
1492 }
1493
1494 for(i=0; i<4; i++){
1495 const int offset= x_offset[i];
1496 const int z0= temp[4*0+i] + temp[4*2+i];
1497 const int z1= temp[4*0+i] - temp[4*2+i];
1498 const int z2= temp[4*1+i] - temp[4*3+i];
1499 const int z3= temp[4*1+i] + temp[4*3+i];
1500
1501 block[stride*0 +offset]= (z0 + z3)>>1;
1502 block[stride*2 +offset]= (z1 + z2)>>1;
1503 block[stride*8 +offset]= (z1 - z2)>>1;
1504 block[stride*10+offset]= (z0 - z3)>>1;
1505 }
1506 }
1507 #endif
1508
1509 #undef xStride
1510 #undef stride
1511
1512 static void chroma_dc_dequant_idct_c(DCTELEM *block, int qp){
1513 const int qmul= dequant_coeff[qp][0];
1514 const int stride= 16*2;
1515 const int xStride= 16;
1516 int a,b,c,d,e;
1517
1518 a= block[stride*0 + xStride*0];
1519 b= block[stride*0 + xStride*1];
1520 c= block[stride*1 + xStride*0];
1521 d= block[stride*1 + xStride*1];
1522
1523 e= a-b;
1524 a= a+b;
1525 b= c-d;
1526 c= c+d;
1527
1528 block[stride*0 + xStride*0]= ((a+c)*qmul + 0)>>1;
1529 block[stride*0 + xStride*1]= ((e+b)*qmul + 0)>>1;
1530 block[stride*1 + xStride*0]= ((a-c)*qmul + 0)>>1;
1531 block[stride*1 + xStride*1]= ((e-b)*qmul + 0)>>1;
1532 }
1533
1534 #if 0
1535 static void chroma_dc_dct_c(DCTELEM *block){
1536 const int stride= 16*2;
1537 const int xStride= 16;
1538 int a,b,c,d,e;
1539
1540 a= block[stride*0 + xStride*0];
1541 b= block[stride*0 + xStride*1];
1542 c= block[stride*1 + xStride*0];
1543 d= block[stride*1 + xStride*1];
1544
1545 e= a-b;
1546 a= a+b;
1547 b= c-d;
1548 c= c+d;
1549
1550 block[stride*0 + xStride*0]= (a+c);
1551 block[stride*0 + xStride*1]= (e+b);
1552 block[stride*1 + xStride*0]= (a-c);
1553 block[stride*1 + xStride*1]= (e-b);
1554 }
1555 #endif
1556
1557 /**
1558 * gets the chroma qp.
1559 */
1560 static inline int get_chroma_qp(H264Context *h, int qscale){
1561
1562 return chroma_qp[clip(qscale + h->pps.chroma_qp_index_offset, 0, 51)];
1563 }
1564
1565
1566 #if 0
1567 static void h264_diff_dct_c(DCTELEM *block, uint8_t *src1, uint8_t *src2, int stride){
1568 int i;
1569 //FIXME try int temp instead of block
1570
1571 for(i=0; i<4; i++){
1572 const int d0= src1[0 + i*stride] - src2[0 + i*stride];
1573 const int d1= src1[1 + i*stride] - src2[1 + i*stride];
1574 const int d2= src1[2 + i*stride] - src2[2 + i*stride];
1575 const int d3= src1[3 + i*stride] - src2[3 + i*stride];
1576 const int z0= d0 + d3;
1577 const int z3= d0 - d3;
1578 const int z1= d1 + d2;
1579 const int z2= d1 - d2;
1580
1581 block[0 + 4*i]= z0 + z1;
1582 block[1 + 4*i]= 2*z3 + z2;
1583 block[2 + 4*i]= z0 - z1;
1584 block[3 + 4*i]= z3 - 2*z2;
1585 }
1586
1587 for(i=0; i<4; i++){
1588 const int z0= block[0*4 + i] + block[3*4 + i];
1589 const int z3= block[0*4 + i] - block[3*4 + i];
1590 const int z1= block[1*4 + i] + block[2*4 + i];
1591 const int z2= block[1*4 + i] - block[2*4 + i];
1592
1593 block[0*4 + i]= z0 + z1;
1594 block[1*4 + i]= 2*z3 + z2;
1595 block[2*4 + i]= z0 - z1;
1596 block[3*4 + i]= z3 - 2*z2;
1597 }
1598 }
1599 #endif
1600
1601 //FIXME need to check that this doesnt overflow signed 32 bit for low qp, iam not sure, its very close
1602 //FIXME check that gcc inlines this (and optimizes intra & seperate_dc stuff away)
1603 static inline int quantize_c(DCTELEM *block, uint8_t *scantable, int qscale, int intra, int seperate_dc){
1604 int i;
1605 const int * const quant_table= quant_coeff[qscale];
1606 const int bias= intra ? (1<<QUANT_SHIFT)/3 : (1<<QUANT_SHIFT)/6;
1607 const unsigned int threshold1= (1<<QUANT_SHIFT) - bias - 1;
1608 const unsigned int threshold2= (threshold1<<1);
1609 int last_non_zero;
1610
1611 if(seperate_dc){
1612 if(qscale<=18){
1613 //avoid overflows
1614 const int dc_bias= intra ? (1<<(QUANT_SHIFT-2))/3 : (1<<(QUANT_SHIFT-2))/6;
1615 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT-2)) - dc_bias - 1;
1616 const unsigned int dc_threshold2= (dc_threshold1<<1);
1617
1618 int level= block[0]*quant_coeff[qscale+18][0];
1619 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1620 if(level>0){
1621 level= (dc_bias + level)>>(QUANT_SHIFT-2);
1622 block[0]= level;
1623 }else{
1624 level= (dc_bias - level)>>(QUANT_SHIFT-2);
1625 block[0]= -level;
1626 }
1627 // last_non_zero = i;
1628 }else{
1629 block[0]=0;
1630 }
1631 }else{
1632 const int dc_bias= intra ? (1<<(QUANT_SHIFT+1))/3 : (1<<(QUANT_SHIFT+1))/6;
1633 const unsigned int dc_threshold1= (1<<(QUANT_SHIFT+1)) - dc_bias - 1;
1634 const unsigned int dc_threshold2= (dc_threshold1<<1);
1635
1636 int level= block[0]*quant_table[0];
1637 if(((unsigned)(level+dc_threshold1))>dc_threshold2){
1638 if(level>0){
1639 level= (dc_bias + level)>>(QUANT_SHIFT+1);
1640 block[0]= level;
1641 }else{
1642 level= (dc_bias - level)>>(QUANT_SHIFT+1);
1643 block[0]= -level;
1644 }
1645 // last_non_zero = i;
1646 }else{
1647 block[0]=0;
1648 }
1649 }
1650 last_non_zero= 0;
1651 i=1;
1652 }else{
1653 last_non_zero= -1;
1654 i=0;
1655 }
1656
1657 for(; i<16; i++){
1658 const int j= scantable[i];
1659 int level= block[j]*quant_table[j];
1660
1661 // if( bias+level >= (1<<(QMAT_SHIFT - 3))
1662 // || bias-level >= (1<<(QMAT_SHIFT - 3))){
1663 if(((unsigned)(level+threshold1))>threshold2){
1664 if(level>0){
1665 level= (bias + level)>>QUANT_SHIFT;
1666 block[j]= level;
1667 }else{
1668 level= (bias - level)>>QUANT_SHIFT;
1669 block[j]= -level;
1670 }
1671 last_non_zero = i;
1672 }else{
1673 block[j]=0;
1674 }
1675 }
1676
1677 return last_non_zero;
1678 }
1679
1680 static void pred4x4_vertical_c(uint8_t *src, uint8_t *topright, int stride){
1681 const uint32_t a= ((uint32_t*)(src-stride))[0];
1682 ((uint32_t*)(src+0*stride))[0]= a;
1683 ((uint32_t*)(src+1*stride))[0]= a;
1684 ((uint32_t*)(src+2*stride))[0]= a;
1685 ((uint32_t*)(src+3*stride))[0]= a;
1686 }
1687
1688 static void pred4x4_horizontal_c(uint8_t *src, uint8_t *topright, int stride){
1689 ((uint32_t*)(src+0*stride))[0]= src[-1+0*stride]*0x01010101;
1690 ((uint32_t*)(src+1*stride))[0]= src[-1+1*stride]*0x01010101;
1691 ((uint32_t*)(src+2*stride))[0]= src[-1+2*stride]*0x01010101;
1692 ((uint32_t*)(src+3*stride))[0]= src[-1+3*stride]*0x01010101;
1693 }
1694
1695 static void pred4x4_dc_c(uint8_t *src, uint8_t *topright, int stride){
1696 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride]
1697 + src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 4) >>3;
1698
1699 ((uint32_t*)(src+0*stride))[0]=
1700 ((uint32_t*)(src+1*stride))[0]=
1701 ((uint32_t*)(src+2*stride))[0]=
1702 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1703 }
1704
1705 static void pred4x4_left_dc_c(uint8_t *src, uint8_t *topright, int stride){
1706 const int dc= ( src[-1+0*stride] + src[-1+1*stride] + src[-1+2*stride] + src[-1+3*stride] + 2) >>2;
1707
1708 ((uint32_t*)(src+0*stride))[0]=
1709 ((uint32_t*)(src+1*stride))[0]=
1710 ((uint32_t*)(src+2*stride))[0]=
1711 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1712 }
1713
1714 static void pred4x4_top_dc_c(uint8_t *src, uint8_t *topright, int stride){
1715 const int dc= ( src[-stride] + src[1-stride] + src[2-stride] + src[3-stride] + 2) >>2;
1716
1717 ((uint32_t*)(src+0*stride))[0]=
1718 ((uint32_t*)(src+1*stride))[0]=
1719 ((uint32_t*)(src+2*stride))[0]=
1720 ((uint32_t*)(src+3*stride))[0]= dc* 0x01010101;
1721 }
1722
1723 static void pred4x4_128_dc_c(uint8_t *src, uint8_t *topright, int stride){
1724 ((uint32_t*)(src+0*stride))[0]=
1725 ((uint32_t*)(src+1*stride))[0]=
1726 ((uint32_t*)(src+2*stride))[0]=
1727 ((uint32_t*)(src+3*stride))[0]= 128U*0x01010101U;
1728 }
1729
1730
1731 #define LOAD_TOP_RIGHT_EDGE\
1732 const int t4= topright[0];\
1733 const int t5= topright[1];\
1734 const int t6= topright[2];\
1735 const int t7= topright[3];\
1736
1737 #define LOAD_LEFT_EDGE\
1738 const int l0= src[-1+0*stride];\
1739 const int l1= src[-1+1*stride];\
1740 const int l2= src[-1+2*stride];\
1741 const int l3= src[-1+3*stride];\
1742
1743 #define LOAD_TOP_EDGE\
1744 const int t0= src[ 0-1*stride];\
1745 const int t1= src[ 1-1*stride];\
1746 const int t2= src[ 2-1*stride];\
1747 const int t3= src[ 3-1*stride];\
1748
1749 static void pred4x4_down_right_c(uint8_t *src, uint8_t *topright, int stride){
1750 const int lt= src[-1-1*stride];
1751 LOAD_TOP_EDGE
1752 LOAD_LEFT_EDGE
1753
1754 src[0+3*stride]=(l3 + 2*l2 + l1 + 2)>>2;
1755 src[0+2*stride]=
1756 src[1+3*stride]=(l2 + 2*l1 + l0 + 2)>>2;
1757 src[0+1*stride]=
1758 src[1+2*stride]=
1759 src[2+3*stride]=(l1 + 2*l0 + lt + 2)>>2;
1760 src[0+0*stride]=
1761 src[1+1*stride]=
1762 src[2+2*stride]=
1763 src[3+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1764 src[1+0*stride]=
1765 src[2+1*stride]=
1766 src[3+2*stride]=(lt + 2*t0 + t1 + 2)>>2;
1767 src[2+0*stride]=
1768 src[3+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1769 src[3+0*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1770 }
1771
1772 static void pred4x4_down_left_c(uint8_t *src, uint8_t *topright, int stride){
1773 LOAD_TOP_EDGE
1774 LOAD_TOP_RIGHT_EDGE
1775 // LOAD_LEFT_EDGE
1776
1777 src[0+0*stride]=(t0 + t2 + 2*t1 + 2)>>2;
1778 src[1+0*stride]=
1779 src[0+1*stride]=(t1 + t3 + 2*t2 + 2)>>2;
1780 src[2+0*stride]=
1781 src[1+1*stride]=
1782 src[0+2*stride]=(t2 + t4 + 2*t3 + 2)>>2;
1783 src[3+0*stride]=
1784 src[2+1*stride]=
1785 src[1+2*stride]=
1786 src[0+3*stride]=(t3 + t5 + 2*t4 + 2)>>2;
1787 src[3+1*stride]=
1788 src[2+2*stride]=
1789 src[1+3*stride]=(t4 + t6 + 2*t5 + 2)>>2;
1790 src[3+2*stride]=
1791 src[2+3*stride]=(t5 + t7 + 2*t6 + 2)>>2;
1792 src[3+3*stride]=(t6 + 3*t7 + 2)>>2;
1793 }
1794
1795 static void pred4x4_vertical_right_c(uint8_t *src, uint8_t *topright, int stride){
1796 const int lt= src[-1-1*stride];
1797 LOAD_TOP_EDGE
1798 LOAD_LEFT_EDGE
1799 const __attribute__((unused)) int unu= l3;
1800
1801 src[0+0*stride]=
1802 src[1+2*stride]=(lt + t0 + 1)>>1;
1803 src[1+0*stride]=
1804 src[2+2*stride]=(t0 + t1 + 1)>>1;
1805 src[2+0*stride]=
1806 src[3+2*stride]=(t1 + t2 + 1)>>1;
1807 src[3+0*stride]=(t2 + t3 + 1)>>1;
1808 src[0+1*stride]=
1809 src[1+3*stride]=(l0 + 2*lt + t0 + 2)>>2;
1810 src[1+1*stride]=
1811 src[2+3*stride]=(lt + 2*t0 + t1 + 2)>>2;
1812 src[2+1*stride]=
1813 src[3+3*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1814 src[3+1*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1815 src[0+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1816 src[0+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1817 }
1818
1819 static void pred4x4_vertical_left_c(uint8_t *src, uint8_t *topright, int stride){
1820 LOAD_TOP_EDGE
1821 LOAD_TOP_RIGHT_EDGE
1822 const __attribute__((unused)) int unu= t7;
1823
1824 src[0+0*stride]=(t0 + t1 + 1)>>1;
1825 src[1+0*stride]=
1826 src[0+2*stride]=(t1 + t2 + 1)>>1;
1827 src[2+0*stride]=
1828 src[1+2*stride]=(t2 + t3 + 1)>>1;
1829 src[3+0*stride]=
1830 src[2+2*stride]=(t3 + t4+ 1)>>1;
1831 src[3+2*stride]=(t4 + t5+ 1)>>1;
1832 src[0+1*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1833 src[1+1*stride]=
1834 src[0+3*stride]=(t1 + 2*t2 + t3 + 2)>>2;
1835 src[2+1*stride]=
1836 src[1+3*stride]=(t2 + 2*t3 + t4 + 2)>>2;
1837 src[3+1*stride]=
1838 src[2+3*stride]=(t3 + 2*t4 + t5 + 2)>>2;
1839 src[3+3*stride]=(t4 + 2*t5 + t6 + 2)>>2;
1840 }
1841
1842 static void pred4x4_horizontal_up_c(uint8_t *src, uint8_t *topright, int stride){
1843 LOAD_LEFT_EDGE
1844
1845 src[0+0*stride]=(l0 + l1 + 1)>>1;
1846 src[1+0*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1847 src[2+0*stride]=
1848 src[0+1*stride]=(l1 + l2 + 1)>>1;
1849 src[3+0*stride]=
1850 src[1+1*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1851 src[2+1*stride]=
1852 src[0+2*stride]=(l2 + l3 + 1)>>1;
1853 src[3+1*stride]=
1854 src[1+2*stride]=(l2 + 2*l3 + l3 + 2)>>2;
1855 src[3+2*stride]=
1856 src[1+3*stride]=
1857 src[0+3*stride]=
1858 src[2+2*stride]=
1859 src[2+3*stride]=
1860 src[3+3*stride]=l3;
1861 }
1862
1863 static void pred4x4_horizontal_down_c(uint8_t *src, uint8_t *topright, int stride){
1864 const int lt= src[-1-1*stride];
1865 LOAD_TOP_EDGE
1866 LOAD_LEFT_EDGE
1867 const __attribute__((unused)) int unu= t3;
1868
1869 src[0+0*stride]=
1870 src[2+1*stride]=(lt + l0 + 1)>>1;
1871 src[1+0*stride]=
1872 src[3+1*stride]=(l0 + 2*lt + t0 + 2)>>2;
1873 src[2+0*stride]=(lt + 2*t0 + t1 + 2)>>2;
1874 src[3+0*stride]=(t0 + 2*t1 + t2 + 2)>>2;
1875 src[0+1*stride]=
1876 src[2+2*stride]=(l0 + l1 + 1)>>1;
1877 src[1+1*stride]=
1878 src[3+2*stride]=(lt + 2*l0 + l1 + 2)>>2;
1879 src[0+2*stride]=
1880 src[2+3*stride]=(l1 + l2+ 1)>>1;
1881 src[1+2*stride]=
1882 src[3+3*stride]=(l0 + 2*l1 + l2 + 2)>>2;
1883 src[0+3*stride]=(l2 + l3 + 1)>>1;
1884 src[1+3*stride]=(l1 + 2*l2 + l3 + 2)>>2;
1885 }
1886
1887 static void pred16x16_vertical_c(uint8_t *src, int stride){
1888 int i;
1889 const uint32_t a= ((uint32_t*)(src-stride))[0];
1890 const uint32_t b= ((uint32_t*)(src-stride))[1];
1891 const uint32_t c= ((uint32_t*)(src-stride))[2];
1892 const uint32_t d= ((uint32_t*)(src-stride))[3];
1893
1894 for(i=0; i<16; i++){
1895 ((uint32_t*)(src+i*stride))[0]= a;
1896 ((uint32_t*)(src+i*stride))[1]= b;
1897 ((uint32_t*)(src+i*stride))[2]= c;
1898 ((uint32_t*)(src+i*stride))[3]= d;
1899 }
1900 }
1901
1902 static void pred16x16_horizontal_c(uint8_t *src, int stride){
1903 int i;
1904
1905 for(i=0; i<16; i++){
1906 ((uint32_t*)(src+i*stride))[0]=
1907 ((uint32_t*)(src+i*stride))[1]=
1908 ((uint32_t*)(src+i*stride))[2]=
1909 ((uint32_t*)(src+i*stride))[3]= src[-1+i*stride]*0x01010101;
1910 }
1911 }
1912
1913 static void pred16x16_dc_c(uint8_t *src, int stride){
1914 int i, dc=0;
1915
1916 for(i=0;i<16; i++){
1917 dc+= src[-1+i*stride];
1918 }
1919
1920 for(i=0;i<16; i++){
1921 dc+= src[i-stride];
1922 }
1923
1924 dc= 0x01010101*((dc + 16)>>5);
1925
1926 for(i=0; i<16; i++){
1927 ((uint32_t*)(src+i*stride))[0]=
1928 ((uint32_t*)(src+i*stride))[1]=
1929 ((uint32_t*)(src+i*stride))[2]=
1930 ((uint32_t*)(src+i*stride))[3]= dc;
1931 }
1932 }
1933
1934 static void pred16x16_left_dc_c(uint8_t *src, int stride){
1935 int i, dc=0;
1936
1937 for(i=0;i<16; i++){
1938 dc+= src[-1+i*stride];
1939 }
1940
1941 dc= 0x01010101*((dc + 8)>>4);
1942
1943 for(i=0; i<16; i++){
1944 ((uint32_t*)(src+i*stride))[0]=
1945 ((uint32_t*)(src+i*stride))[1]=
1946 ((uint32_t*)(src+i*stride))[2]=
1947 ((uint32_t*)(src+i*stride))[3]= dc;
1948 }
1949 }
1950
1951 static void pred16x16_top_dc_c(uint8_t *src, int stride){
1952 int i, dc=0;
1953
1954 for(i=0;i<16; i++){
1955 dc+= src[i-stride];
1956 }
1957 dc= 0x01010101*((dc + 8)>>4);
1958
1959 for(i=0; i<16; i++){
1960 ((uint32_t*)(src+i*stride))[0]=
1961 ((uint32_t*)(src+i*stride))[1]=
1962 ((uint32_t*)(src+i*stride))[2]=
1963 ((uint32_t*)(src+i*stride))[3]= dc;
1964 }
1965 }
1966
1967 static void pred16x16_128_dc_c(uint8_t *src, int stride){
1968 int i;
1969
1970 for(i=0; i<16; i++){
1971 ((uint32_t*)(src+i*stride))[0]=
1972 ((uint32_t*)(src+i*stride))[1]=
1973 ((uint32_t*)(src+i*stride))[2]=
1974 ((uint32_t*)(src+i*stride))[3]= 0x01010101U*128U;
1975 }
1976 }
1977
1978 static inline void pred16x16_plane_compat_c(uint8_t *src, int stride, const int svq3){
1979 int i, j, k;
1980 int a;
1981 uint8_t *cm = cropTbl + MAX_NEG_CROP;
1982 const uint8_t * const src0 = src+7-stride;
1983 const uint8_t *src1 = src+8*stride-1;
1984 const uint8_t *src2 = src1-2*stride; // == src+6*stride-1;
1985 int H = src0[1] - src0[-1];
1986 int V = src1[0] - src2[ 0];
1987 for(k=2; k<=8; ++k) {
1988 src1 += stride; src2 -= stride;
1989 H += k*(src0[k] - src0[-k]);
1990 V += k*(src1[0] - src2[ 0]);
1991 }
1992 if(svq3){
1993 H = ( 5*(H/4) ) / 16;
1994 V = ( 5*(V/4) ) / 16;
1995
1996 /* required for 100% accuracy */
1997 i = H; H = V; V = i;
1998 }else{
1999 H = ( 5*H+32 ) >> 6;
2000 V = ( 5*V+32 ) >> 6;
2001 }
2002
2003 a = 16*(src1[0] + src2[16] + 1) - 7*(V+H);
2004 for(j=16; j>0; --j) {
2005 int b = a;
2006 a += V;
2007 for(i=-16; i<0; i+=4) {
2008 src[16+i] = cm[ (b ) >> 5 ];
2009 src[17+i] = cm[ (b+ H) >> 5 ];
2010 src[18+i] = cm[ (b+2*H) >> 5 ];
2011 src[19+i] = cm[ (b+3*H) >> 5 ];
2012 b += 4*H;
2013 }
2014 src += stride;
2015 }
2016 }
2017
2018 static void pred16x16_plane_c(uint8_t *src, int stride){
2019 pred16x16_plane_compat_c(src, stride, 0);
2020 }
2021
2022 static void pred8x8_vertical_c(uint8_t *src, int stride){
2023 int i;
2024 const uint32_t a= ((uint32_t*)(src-stride))[0];
2025 const uint32_t b= ((uint32_t*)(src-stride))[1];
2026
2027 for(i=0; i<8; i++){
2028 ((uint32_t*)(src+i*stride))[0]= a;
2029 ((uint32_t*)(src+i*stride))[1]= b;
2030 }
2031 }
2032
2033 static void pred8x8_horizontal_c(uint8_t *src, int stride){
2034 int i;
2035
2036 for(i=0; i<8; i++){
2037 ((uint32_t*)(src+i*stride))[0]=
2038 ((uint32_t*)(src+i*stride))[1]= src[-1+i*stride]*0x01010101;
2039 }
2040 }
2041
2042 static void pred8x8_128_dc_c(uint8_t *src, int stride){
2043 int i;
2044
2045 for(i=0; i<4; i++){
2046 ((uint32_t*)(src+i*stride))[0]=
2047 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
2048 }
2049 for(i=4; i<8; i++){
2050 ((uint32_t*)(src+i*stride))[0]=
2051 ((uint32_t*)(src+i*stride))[1]= 0x01010101U*128U;
2052 }
2053 }
2054
2055 static void pred8x8_left_dc_c(uint8_t *src, int stride){
2056 int i;
2057 int dc0, dc2;
2058
2059 dc0=dc2=0;
2060 for(i=0;i<4; i++){
2061 dc0+= src[-1+i*stride];
2062 dc2+= src[-1+(i+4)*stride];
2063 }
2064 dc0= 0x01010101*((dc0 + 2)>>2);
2065 dc2= 0x01010101*((dc2 + 2)>>2);
2066
2067 for(i=0; i<4; i++){
2068 ((uint32_t*)(src+i*stride))[0]=
2069 ((uint32_t*)(src+i*stride))[1]= dc0;
2070 }
2071 for(i=4; i<8; i++){
2072 ((uint32_t*)(src+i*stride))[0]=
2073 ((uint32_t*)(src+i*stride))[1]= dc2;
2074 }
2075 }
2076
2077 static void pred8x8_top_dc_c(uint8_t *src, int stride){
2078 int i;
2079 int dc0, dc1;
2080
2081 dc0=dc1=0;
2082 for(i=0;i<4; i++){
2083 dc0+= src[i-stride];
2084 dc1+= src[4+i-stride];
2085 }
2086 dc0= 0x01010101*((dc0 + 2)>>2);
2087 dc1= 0x01010101*((dc1 + 2)>>2);
2088
2089 for(i=0; i<4; i++){
2090 ((uint32_t*)(src+i*stride))[0]= dc0;
2091 ((uint32_t*)(src+i*stride))[1]= dc1;
2092 }
2093 for(i=4; i<8; i++){
2094 ((uint32_t*)(src+i*stride))[0]= dc0;
2095 ((uint32_t*)(src+i*stride))[1]= dc1;
2096 }
2097 }
2098
2099
2100 static void pred8x8_dc_c(uint8_t *src, int stride){
2101 int i;
2102 int dc0, dc1, dc2, dc3;
2103
2104 dc0=dc1=dc2=0;
2105 for(i=0;i<4; i++){
2106 dc0+= src[-1+i*stride] + src[i-stride];
2107 dc1+= src[4+i-stride];
2108 dc2+= src[-1+(i+4)*stride];
2109 }
2110 dc3= 0x01010101*((dc1 + dc2 + 4)>>3);
2111 dc0= 0x01010101*((dc0 + 4)>>3);
2112 dc1= 0x01010101*((dc1 + 2)>>2);
2113 dc2= 0x01010101*((dc2 + 2)>>2);
2114
2115 for(i=0; i<4; i++){
2116 ((uint32_t*)(src+i*stride))[0]= dc0;
2117 ((uint32_t*)(src+i*stride))[1]= dc1;
2118 }
2119 for(i=4; i<8; i++){
2120 ((uint32_t*)(src+i*stride))[0]= dc2;
2121 ((uint32_t*)(src+i*stride))[1]= dc3;
2122 }
2123 }
2124
2125 static void pred8x8_plane_c(uint8_t *src, int stride){
2126 int j, k;
2127 int a;
2128 uint8_t *cm = cropTbl + MAX_NEG_CROP;
2129 const uint8_t * const src0 = src+3-stride;
2130 const uint8_t *src1 = src+4*stride-1;
2131 const uint8_t *src2 = src1-2*stride; // == src+2*stride-1;
2132 int H = src0[1] - src0[-1];
2133 int V = src1[0] - src2[ 0];
2134 for(k=2; k<=4; ++k) {
2135 src1 += stride; src2 -= stride;
2136 H += k*(src0[k] - src0[-k]);
2137 V += k*(src1[0] - src2[ 0]);
2138 }
2139 H = ( 17*H+16 ) >> 5;
2140 V = ( 17*V+16 ) >> 5;
2141
2142 a = 16*(src1[0] + src2[8]+1) - 3*(V+H);
2143 for(j=8; j>0; --j) {
2144 int b = a;
2145 a += V;
2146 src[0] = cm[ (b ) >> 5 ];
2147 src[1] = cm[ (b+ H) >> 5 ];
2148 src[2] = cm[ (b+2*H) >> 5 ];
2149 src[3] = cm[ (b+3*H) >> 5 ];
2150 src[4] = cm[ (b+4*H) >> 5 ];
2151 src[5] = cm[ (b+5*H) >> 5 ];
2152 src[6] = cm[ (b+6*H) >> 5 ];
2153 src[7] = cm[ (b+7*H) >> 5 ];
2154 src += stride;
2155 }
2156 }
2157
2158 static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
2159 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2160 int src_x_offset, int src_y_offset,
2161 qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op){
2162 MpegEncContext * const s = &h->s;
2163 const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
2164 const int my= h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
2165 const int luma_xy= (mx&3) + ((my&3)<<2);
2166 uint8_t * src_y = pic->data[0] + (mx>>2) + (my>>2)*s->linesize;
2167 uint8_t * src_cb= pic->data[1] + (mx>>3) + (my>>3)*s->uvlinesize;
2168 uint8_t * src_cr= pic->data[2] + (mx>>3) + (my>>3)*s->uvlinesize;
2169 int extra_width= (s->flags&CODEC_FLAG_EMU_EDGE) ? 0 : 16; //FIXME increase edge?, IMHO not worth it
2170 int extra_height= extra_width;
2171 int emu=0;
2172 const int full_mx= mx>>2;
2173 const int full_my= my>>2;
2174
2175 assert(pic->data[0]);
2176
2177 if(mx&7) extra_width -= 3;
2178 if(my&7) extra_height -= 3;
2179
2180 if( full_mx < 0-extra_width
2181 || full_my < 0-extra_height
2182 || full_mx + 16/*FIXME*/ > s->width + extra_width
2183 || full_my + 16/*FIXME*/ > s->height + extra_height){
2184 ff_emulated_edge_mc(s->edge_emu_buffer, src_y - 2 - 2*s->linesize, s->linesize, 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, s->width, s->height);
2185 src_y= s->edge_emu_buffer + 2 + 2*s->linesize;
2186 emu=1;
2187 }
2188
2189 qpix_op[luma_xy](dest_y, src_y, s->linesize); //FIXME try variable height perhaps?
2190 if(!square){
2191 qpix_op[luma_xy](dest_y + delta, src_y + delta, s->linesize);
2192 }
2193
2194 if(s->flags&CODEC_FLAG_GRAY) return;
2195
2196 if(emu){
2197 ff_emulated_edge_mc(s->edge_emu_buffer, src_cb, s->uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), s->width>>1, s->height>>1);
2198 src_cb= s->edge_emu_buffer;
2199 }
2200 chroma_op(dest_cb, src_cb, s->uvlinesize, chroma_height, mx&7, my&7);
2201
2202 if(emu){
2203 ff_emulated_edge_mc(s->edge_emu_buffer, src_cr, s->uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), s->width>>1, s->height>>1);
2204 src_cr= s->edge_emu_buffer;
2205 }
2206 chroma_op(dest_cr, src_cr, s->uvlinesize, chroma_height, mx&7, my&7);
2207 }
2208
2209 static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
2210 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2211 int x_offset, int y_offset,
2212 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2213 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2214 int list0, int list1){
2215 MpegEncContext * const s = &h->s;
2216 qpel_mc_func *qpix_op= qpix_put;
2217 h264_chroma_mc_func chroma_op= chroma_put;
2218
2219 dest_y += 2*x_offset + 2*y_offset*s-> linesize;
2220 dest_cb += x_offset + y_offset*s->uvlinesize;
2221 dest_cr += x_offset + y_offset*s->uvlinesize;
2222 x_offset += 8*s->mb_x;
2223 y_offset += 8*s->mb_y;
2224
2225 if(list0){
2226 Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
2227 mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
2228 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2229 qpix_op, chroma_op);
2230
2231 qpix_op= qpix_avg;
2232 chroma_op= chroma_avg;
2233 }
2234
2235 if(list1){
2236 Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
2237 mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
2238 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2239 qpix_op, chroma_op);
2240 }
2241 }
2242
2243 static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
2244 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2245 int x_offset, int y_offset,
2246 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2247 h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
2248 h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
2249 int list0, int list1){
2250 MpegEncContext * const s = &h->s;
2251
2252 dest_y += 2*x_offset + 2*y_offset*s-> linesize;
2253 dest_cb += x_offset + y_offset*s->uvlinesize;
2254 dest_cr += x_offset + y_offset*s->uvlinesize;
2255 x_offset += 8*s->mb_x;
2256 y_offset += 8*s->mb_y;
2257
2258 if(list0 && list1){
2259 /* don't optimize for luma-only case, since B-frames usually
2260 * use implicit weights => chroma too. */
2261 uint8_t *tmp_cb = s->obmc_scratchpad;
2262 uint8_t *tmp_cr = tmp_cb + 8*s->uvlinesize;
2263 uint8_t *tmp_y = tmp_cr + 8*s->uvlinesize;
2264 int refn0 = h->ref_cache[0][ scan8[n] ];
2265 int refn1 = h->ref_cache[1][ scan8[n] ];
2266
2267 mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
2268 dest_y, dest_cb, dest_cr,
2269 x_offset, y_offset, qpix_put, chroma_put);
2270 mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
2271 tmp_y, tmp_cb, tmp_cr,
2272 x_offset, y_offset, qpix_put, chroma_put);
2273
2274 if(h->use_weight == 2){
2275 int weight0 = h->implicit_weight[refn0][refn1];
2276 int weight1 = 64 - weight0;
2277 luma_weight_avg( dest_y, tmp_y, s-> linesize, 5, weight0, weight1, 0, 0);
2278 chroma_weight_avg(dest_cb, tmp_cb, s->uvlinesize, 5, weight0, weight1, 0, 0);
2279 chroma_weight_avg(dest_cr, tmp_cr, s->uvlinesize, 5, weight0, weight1, 0, 0);
2280 }else{
2281 luma_weight_avg(dest_y, tmp_y, s->linesize, h->luma_log2_weight_denom,
2282 h->luma_weight[0][refn0], h->luma_weight[1][refn1],
2283 h->luma_offset[0][refn0], h->luma_offset[1][refn1]);
2284 chroma_weight_avg(dest_cb, tmp_cb, s->uvlinesize, h->chroma_log2_weight_denom,
2285 h->chroma_weight[0][refn0][0], h->chroma_weight[1][refn1][0],
2286 h->chroma_offset[0][refn0][0], h->chroma_offset[1][refn1][0]);
2287 chroma_weight_avg(dest_cr, tmp_cr, s->uvlinesize, h->chroma_log2_weight_denom,
2288 h->chroma_weight[0][refn0][1], h->chroma_weight[1][refn1][1],
2289 h->chroma_offset[0][refn0][1], h->chroma_offset[1][refn1][1]);
2290 }
2291 }else{
2292 int list = list1 ? 1 : 0;
2293 int refn = h->ref_cache[list][ scan8[n] ];
2294 Picture *ref= &h->ref_list[list][refn];
2295 mc_dir_part(h, ref, n, square, chroma_height, delta, list,
2296 dest_y, dest_cb, dest_cr, x_offset, y_offset,
2297 qpix_put, chroma_put);
2298
2299 luma_weight_op(dest_y, s->linesize, h->luma_log2_weight_denom,
2300 h->luma_weight[list][refn], h->luma_offset[list][refn]);
2301 if(h->use_weight_chroma){
2302 chroma_weight_op(dest_cb, s->uvlinesize, h->chroma_log2_weight_denom,
2303 h->chroma_weight[list][refn][0], h->chroma_offset[list][refn][0]);
2304 chroma_weight_op(dest_cr, s->uvlinesize, h->chroma_log2_weight_denom,
2305 h->chroma_weight[list][refn][1], h->chroma_offset[list][refn][1]);
2306 }
2307 }
2308 }
2309
2310 static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
2311 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2312 int x_offset, int y_offset,
2313 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
2314 qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
2315 h264_weight_func *weight_op, h264_biweight_func *weight_avg,
2316 int list0, int list1){
2317 if((h->use_weight==2 && list0 && list1
2318 && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ] != 32))
2319 || h->use_weight==1)
2320 mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2321 x_offset, y_offset, qpix_put, chroma_put,
2322 weight_op[0], weight_op[3], weight_avg[0], weight_avg[3], list0, list1);
2323 else
2324 mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
2325 x_offset, y_offset, qpix_put, chroma_put, qpix_avg, chroma_avg, list0, list1);
2326 }
2327
2328 static void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
2329 qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
2330 qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
2331 h264_weight_func *weight_op, h264_biweight_func *weight_avg){
2332 MpegEncContext * const s = &h->s;
2333 const int mb_xy= s->mb_x + s->mb_y*s->mb_stride;
2334 const int mb_type= s->current_picture.mb_type[mb_xy];
2335
2336 assert(IS_INTER(mb_type));
2337
2338 if(IS_16X16(mb_type)){
2339 mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
2340 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
2341 &weight_op[0], &weight_avg[0],
2342 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2343 }else if(IS_16X8(mb_type)){
2344 mc_part(h, 0, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 0,
2345 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2346 &weight_op[1], &weight_avg[1],
2347 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2348 mc_part(h, 8, 0, 4, 8, dest_y, dest_cb, dest_cr, 0, 4,
2349 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
2350 &weight_op[1], &weight_avg[1],
2351 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2352 }else if(IS_8X16(mb_type)){
2353 mc_part(h, 0, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 0, 0,
2354 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2355 &weight_op[2], &weight_avg[2],
2356 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1));
2357 mc_part(h, 4, 0, 8, 8*s->linesize, dest_y, dest_cb, dest_cr, 4, 0,
2358 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2359 &weight_op[2], &weight_avg[2],
2360 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1));
2361 }else{
2362 int i;
2363
2364 assert(IS_8X8(mb_type));
2365
2366 for(i=0; i<4; i++){
2367 const int sub_mb_type= h->sub_mb_type[i];
2368 const int n= 4*i;
2369 int x_offset= (i&1)<<2;
2370 int y_offset= (i&2)<<1;
2371
2372 if(IS_SUB_8X8(sub_mb_type)){
2373 mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2374 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
2375 &weight_op[3], &weight_avg[3],
2376 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2377 }else if(IS_SUB_8X4(sub_mb_type)){
2378 mc_part(h, n , 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2379 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2380 &weight_op[4], &weight_avg[4],
2381 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2382 mc_part(h, n+2, 0, 2, 4, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
2383 qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
2384 &weight_op[4], &weight_avg[4],
2385 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2386 }else if(IS_SUB_4X8(sub_mb_type)){
2387 mc_part(h, n , 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
2388 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2389 &weight_op[5], &weight_avg[5],
2390 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2391 mc_part(h, n+1, 0, 4, 4*s->linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
2392 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2393 &weight_op[5], &weight_avg[5],
2394 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2395 }else{
2396 int j;
2397 assert(IS_SUB_4X4(sub_mb_type));
2398 for(j=0; j<4; j++){
2399 int sub_x_offset= x_offset + 2*(j&1);
2400 int sub_y_offset= y_offset + (j&2);
2401 mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
2402 qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
2403 &weight_op[6], &weight_avg[6],
2404 IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1));
2405 }
2406 }
2407 }
2408 }
2409 }
2410
2411 static void decode_init_vlc(H264Context *h){
2412 static int done = 0;
2413
2414 if (!done) {
2415 int i;
2416 done = 1;
2417
2418 init_vlc(&chroma_dc_coeff_token_vlc, CHROMA_DC_COEFF_TOKEN_VLC_BITS, 4*5,
2419 &chroma_dc_coeff_token_len [0], 1, 1,
2420 &chroma_dc_coeff_token_bits[0], 1, 1, 1);
2421
2422 for(i=0; i<4; i++){
2423 init_vlc(&coeff_token_vlc[i], COEFF_TOKEN_VLC_BITS, 4*17,
2424 &coeff_token_len [i][0], 1, 1,
2425 &coeff_token_bits[i][0], 1, 1, 1);
2426 }
2427
2428 for(i=0; i<3; i++){
2429 init_vlc(&chroma_dc_total_zeros_vlc[i], CHROMA_DC_TOTAL_ZEROS_VLC_BITS, 4,
2430 &chroma_dc_total_zeros_len [i][0], 1, 1,
2431 &chroma_dc_total_zeros_bits[i][0], 1, 1, 1);
2432 }
2433 for(i=0; i<15; i++){
2434 init_vlc(&total_zeros_vlc[i], TOTAL_ZEROS_VLC_BITS, 16,
2435 &total_zeros_len [i][0], 1, 1,
2436 &total_zeros_bits[i][0], 1, 1, 1);
2437 }
2438
2439 for(i=0; i<6; i++){
2440 init_vlc(&run_vlc[i], RUN_VLC_BITS, 7,
2441 &run_len [i][0], 1, 1,
2442 &run_bits[i][0], 1, 1, 1);
2443 }
2444 init_vlc(&run7_vlc, RUN7_VLC_BITS, 16,
2445 &run_len [6][0], 1, 1,
2446 &run_bits[6][0], 1, 1, 1);
2447 }
2448 }
2449
2450 /**
2451 * Sets the intra prediction function pointers.
2452 */
2453 static void init_pred_ptrs(H264Context *h){
2454 // MpegEncContext * const s = &h->s;
2455
2456 h->pred4x4[VERT_PRED ]= pred4x4_vertical_c;
2457 h->pred4x4[HOR_PRED ]= pred4x4_horizontal_c;
2458 h->pred4x4[DC_PRED ]= pred4x4_dc_c;
2459 h->pred4x4[DIAG_DOWN_LEFT_PRED ]= pred4x4_down_left_c;
2460 h->pred4x4[DIAG_DOWN_RIGHT_PRED]= pred4x4_down_right_c;
2461 h->pred4x4[VERT_RIGHT_PRED ]= pred4x4_vertical_right_c;
2462 h->pred4x4[HOR_DOWN_PRED ]= pred4x4_horizontal_down_c;
2463 h->pred4x4[VERT_LEFT_PRED ]= pred4x4_vertical_left_c;
2464 h->pred4x4[HOR_UP_PRED ]= pred4x4_horizontal_up_c;
2465 h->pred4x4[LEFT_DC_PRED ]= pred4x4_left_dc_c;
2466 h->pred4x4[TOP_DC_PRED ]= pred4x4_top_dc_c;
2467 h->pred4x4[DC_128_PRED ]= pred4x4_128_dc_c;
2468
2469 h->pred8x8[DC_PRED8x8 ]= pred8x8_dc_c;
2470 h->pred8x8[VERT_PRED8x8 ]= pred8x8_vertical_c;
2471 h->pred8x8[HOR_PRED8x8 ]= pred8x8_horizontal_c;
2472 h->pred8x8[PLANE_PRED8x8 ]= pred8x8_plane_c;
2473 h->pred8x8[LEFT_DC_PRED8x8]= pred8x8_left_dc_c;
2474 h->pred8x8[TOP_DC_PRED8x8 ]= pred8x8_top_dc_c;
2475 h->pred8x8[DC_128_PRED8x8 ]= pred8x8_128_dc_c;
2476
2477 h->pred16x16[DC_PRED8x8 ]= pred16x16_dc_c;
2478 h->pred16x16[VERT_PRED8x8 ]= pred16x16_vertical_c;
2479 h->pred16x16[HOR_PRED8x8 ]= pred16x16_horizontal_c;
2480 h->pred16x16[PLANE_PRED8x8 ]= pred16x16_plane_c;
2481 h->pred16x16[LEFT_DC_PRED8x8]= pred16x16_left_dc_c;
2482 h->pred16x16[TOP_DC_PRED8x8 ]= pred16x16_top_dc_c;
2483 h->pred16x16[DC_128_PRED8x8 ]= pred16x16_128_dc_c;
2484 }
2485
2486 static void free_tables(H264Context *h){
2487 av_freep(&h->intra4x4_pred_mode);
2488 av_freep(&h->chroma_pred_mode_table);
2489 av_freep(&h->cbp_table);
2490 av_freep(&h->mvd_table[0]);
2491 av_freep(&h->mvd_table[1]);
2492 av_freep(&h->direct_table);
2493 av_freep(&h->non_zero_count);
2494 av_freep(&h->slice_table_base);
2495 av_freep(&h->top_border);
2496 h->slice_table= NULL;
2497
2498 av_freep(&h->mb2b_xy);
2499 av_freep(&h->mb2b8_xy);
2500
2501 av_freep(&h->s.obmc_scratchpad);
2502 }
2503
2504 /**
2505 * allocates tables.
2506 * needs widzh/height
2507 */
2508 static int alloc_tables(H264Context *h){
2509 MpegEncContext * const s = &h->s;
2510 const int big_mb_num= s->mb_stride * (s->mb_height+1);
2511 int x,y;
2512
2513 CHECKED_ALLOCZ(h->intra4x4_pred_mode, big_mb_num * 8 * sizeof(uint8_t))
2514
2515 CHECKED_ALLOCZ(h->non_zero_count , big_mb_num * 16 * sizeof(uint8_t))
2516 CHECKED_ALLOCZ(h->slice_table_base , big_mb_num * sizeof(uint8_t))
2517 CHECKED_ALLOCZ(h->top_border , s->mb_width * (16+8+8) * sizeof(uint8_t))
2518 CHECKED_ALLOCZ(h->cbp_table, big_mb_num * sizeof(uint16_t))
2519
2520 if( h->pps.cabac ) {
2521 CHECKED_ALLOCZ(h->chroma_pred_mode_table, big_mb_num * sizeof(uint8_t))
2522 CHECKED_ALLOCZ(h->mvd_table[0], 32*big_mb_num * sizeof(uint16_t));
2523 CHECKED_ALLOCZ(h->mvd_table[1], 32*big_mb_num * sizeof(uint16_t));
2524 CHECKED_ALLOCZ(h->direct_table, 32*big_mb_num * sizeof(uint8_t));
2525 }
2526
2527 memset(h->slice_table_base, -1, big_mb_num * sizeof(uint8_t));
2528 h->slice_table= h->slice_table_base + s->mb_stride + 1;
2529
2530 CHECKED_ALLOCZ(h->mb2b_xy , big_mb_num * sizeof(uint16_t));
2531 CHECKED_ALLOCZ(h->mb2b8_xy , big_mb_num * sizeof(uint16_t));
2532 for(y=0; y<s->mb_height; y++){
2533 for(x=0; x<s->mb_width; x++){
2534 const int mb_xy= x + y*s->mb_stride;
2535 const int b_xy = 4*x + 4*y*h->b_stride;
2536 const int b8_xy= 2*x + 2*y*h->b8_stride;
2537
2538 h->mb2b_xy [mb_xy]= b_xy;
2539 h->mb2b8_xy[mb_xy]= b8_xy;
2540 }
2541 }
2542
2543 CHECKED_ALLOCZ(s->obmc_scratchpad, 16*s->linesize + 2*8*s->uvlinesize);
2544
2545 return 0;
2546 fail:
2547 free_tables(h);
2548 return -1;
2549 }
2550
2551 static void common_init(H264Context *h){
2552 MpegEncContext * const s = &h->s;
2553
2554 s->width = s->avctx->width;
2555 s->height = s->avctx->height;
2556 s->codec_id= s->avctx->codec->id;
2557
2558 init_pred_ptrs(h);
2559
2560 s->unrestricted_mv=1;
2561 s->decode=1; //FIXME
2562 }
2563
2564 static int decode_init(AVCodecContext *avctx){
2565 H264Context *h= avctx->priv_data;
2566 MpegEncContext * const s = &h->s;
2567
2568 MPV_decode_defaults(s);
2569
2570 s->avctx = avctx;
2571 common_init(h);
2572
2573 s->out_format = FMT_H264;
2574 s->workaround_bugs= avctx->workaround_bugs;
2575
2576 // set defaults
2577 // s->decode_mb= ff_h263_decode_mb;
2578 s->low_delay= 1;
2579 avctx->pix_fmt= PIX_FMT_YUV420P;
2580
2581 decode_init_vlc(h);
2582
2583 if(avctx->codec_tag != 0x31637661 && avctx->codec_tag != 0x31435641) // avc1
2584 h->is_avc = 0;
2585 else {
2586 if((avctx->extradata_size == 0) || (avctx->extradata == NULL)) {
2587 av_log(avctx, AV_LOG_ERROR, "AVC codec requires avcC data\n");
2588 return -1;
2589 }
2590 h->is_avc = 1;
2591 h->got_avcC = 0;
2592 }
2593
2594 return 0;
2595 }
2596
2597 static void frame_start(H264Context *h){
2598 MpegEncContext * const s = &h->s;
2599 int i;
2600
2601 MPV_frame_start(s, s->avctx);
2602 ff_er_frame_start(s);
2603 h->mmco_index=0;
2604
2605 assert(s->linesize && s->uvlinesize);
2606
2607 for(i=0; i<16; i++){
2608 h->block_offset[i]= 4*((scan8[i] - scan8[0])&7) + 4*s->linesize*((scan8[i] - scan8[0])>>3);
2609 h->chroma_subblock_offset[i]= 2*((scan8[i] - scan8[0])&7) + 2*s->uvlinesize*((scan8[i] - scan8[0])>>3);
2610 }
2611 for(i=0; i<4; i++){
2612 h->block_offset[16+i]=
2613 h->block_offset[20+i]= 4*((scan8[i] - scan8[0])&7) + 4*s->uvlinesize*((scan8[i] - scan8[0])>>3);